Moonshot AI's Kimi K3 challenges US rivals as Chinese AI closes the performance gap

Does China's latest AI model finally equal US rivals? What scientists think

The latest Chinese-built large language model (LLM) is impressing scientists with its size and capabilities. Last week, Beijing-based Moonshot AI unveiled 'Kimi K3', a powerful reasoning LLM that can handle large swathes of text. The company's own tests found that K3 can match or outperform rival US models on tasks such as coding and manipulating spreadsheets. Not since DeepSeek has a Chinese artificial-intelligence model caused quite so much buzz. "It is a turning point," says Joel Pearson, a cognitive neuroscientist at the University of New South Wales, Sydney, Australia, who studies how AI is impacting people's lives. "People are calling it a 'Sputnik moment'," he adds. Three days after the model's release, Moonshot AI, said it had paused new sign-ups to K3 because demand had pushed its system close to the limits of its processing capacity. The model's launch on 16 July came just before the 2026 Artificial Intelligence World Conference in Shanghai. The conference began with Chinese President Xi Jinping announcing the formal start of a global alliance to create regulation that ensures AI is safe and benefits people. "In China's view, all countries should take a people-centred approach and develop AI for the positive and for good," Xi said at the conference, adding that AI should be a driver for "shared prosperity and common security". The timing of Kimi K3's release is significant, says Mehwish Nasim, an AI researcher at the University of Western Australia in Perth. The model was launched just weeks after Claude Fable 5 by US company Anthropic, and days after OpenAI released its latest GPT-5.6 model. "China is signalling its ambition not only to build frontier AI systems but also to help shape the international AI ecosystem and its governance," she says. Largest open-weight model K3 is open weight, like its predecessor Kimi K2 and DeepSeek models, which were developed by a Hangzhou-based firm of the same name, meaning its core components are publicly available and can be downloaded and modified by researchers for free, while information on the model's training is private. It can also be accessed using an 'application programming interface', which allows software to communicate, at a lower cost than proprietary LLMs, such as those owned by OpenAI, Google and Anthropic. These LLMs are closed weight, meaning they cannot be downloaded and modified. Moonshot says that the weights, or parameters, will be released on 27 July. Rahul Shome, a robotics and AI researcher at the Australian National University in Canberra, says that there has been a push from consumers, developers and researchers to make AI models open weight. K3 is particularly exciting because it would be the largest open-weight model, says Shome, with 2.8 trillion parameters and a working memory of one million tokens (the units of text used by AI models). The model is too large to run on personal devices, however, so it will probably require significant investment by institutions to run it, he adds. K3's large working memory means it could remember the contents of thousands of lines of code or a whole book and reduce the likelihood of producing fabricated information called hallucination, says Niusha Shafiabady, a computational intelligence researcher at Australian Catholic University in Sydney. Shafiabady says that she is going to test K3 for herself by asking it to summarize research findings. Closing the gap The performance and capabilities of open-weight models have historically lagged behind those of US proprietary ones. But the gap seems to be the tightest it has ever been, says Aaron Snoswell, an AI accountability researcher at the Queensland University of Technology Generative AI Lab, Brisbane. Pearson says that Chinese open-weight models are changing how investors and users perceive the value of US frontier models. Google, Anthropic and OpenAI spend billions of dollars training their models compared with the much lower budgets of Chinese companies. Moreover, users could lose access to US models if the government decides to impose restrictions, says Pearson. Last month, Anthropic temporarily disabled its most advanced models, Fable 5 and Mythos 5, for all users after it was directed by the US government to suspend access for foreign nationals. Access to Fable 5 was restored this month, but Mythos 5 is only available to specific US organizations. Toby Walsh, a computer scientist at the University of New South Wales, says that Chinese AI companies have been able to build frontier AI models despite the United States restricting China's access to advanced AI chips, citing national-security concerns. "One suspects that necessity here was the mother of invention," he adds.

Arcee, a US open source AI lab, says Chinese models are not inherently dangerous

As Chinese open-weight AI models grow in capability and popularity, arguments about what should be done about them have once again reached a fever pitch. There's talk that the Trump administration might try to ban them (though it hasn't yet acted on the idea). Meanwhile, proprietary model makers, particularly OpenAI and Anthropic, appear increasingly concerned about them. Open-weight models such as Moonshot AI's Kimi K3 or Alibaba's Qwen offer inference at a fraction of the token cost of closed-source models from these large U.S. labs. The fear is that they also pose some sort of threat. Certainly, they threaten the profit margins of the large proprietary AI labs. But should enterprises running these models in their own data centers succumb to the fear that they could be a vector for Chinese hackers? No, says Lucas Atkins, the CTO of Arcee, which is building open models to give U.S. companies a homegrown alternative to Chinese models. If any startup would benefit from a ban on Chinese models, Arcee would. But Atkins says China's open models are no more dangerous than any other open-source software a company may use. In fact, he says, they even offer benefits even to his own company. "A lot of people view this as similar to a Chinese software program. Like, it was coded with these x, y, z intentions" that a bad actor could simply command, he said. "That is fundamentally not how these models are trained. There is really not any way for an Arcee, or an Alibaba, to make a model, have someone run it in their own environment and for us have any access to it whatsoever," he explained. While most of these models are what's known as "open weight," and are not really fully open-source software, the source code (the part that will actually run on servers), if it is downloaded from open-source sites like Hugging Face, is similarly largely visible and reviewable. (What isn't available is the methods and data used to train the models.) Large organizations should put any model core through their security testing and inspection processes, and they will also often post-train the models for their specific uses. So they work with, optimize, and understand the models before people start sending them prompts. Could a model that is used for coding somehow throw malicious backdoors into the code it writes? Again, while that's theoretically possible, it would require acrobatic feats to accomplish. "There's no reason that a sophisticated enough actor couldn't train a model to be a completely amazing coding model in every circumstance, but when presented with a certain type of code base ... some hidden training would kick in," Atkins, who spends his days training models, postulated. But he adds: "I don't know how you would do this." Because large language models are by nature creative, the odds are slim of getting a contemporary model to spit out malware in response to a preplanned perfect storm of context and prompt. Even slimmer are the chances that any enterprise would then use that code. Could it happen in the future? That's anyone's guess. But enterprises are also building their AI apps to be model-agnostic and to use multiple models. So even if Chinese models are the best for the price today, enterprises won't be locked into using them forever. "I think instead of the conversation being about how to ban Chinese models, it should be about how do we foster a good, open ecosystem here in the U.S.," Atkins says. Arcee also gains advantages from Chinese models. Because they are open, the startup "benefits from those models being good because we can learn what they did. We can build on top of them. Then they can learn what we do," he says. "We have tremendous respect for the people building those models, the individual researchers." Ultimately, the way to compete with Chinese models "is to release a model that is better," says Atkins. "We need to give them something to talk about."

China's Open AI Models Are Challenging Silicon Valley's Playbook

The AI industry is not quite experiencing a DeepSeek 2.0 moment, but it feels very close. The leading Chinese AI labs have been on a roll lately, releasing a series of almost cutting-edge open-source models. Z.ai released GLM 5.2 in June, Moonshot AI released Kimi K3 last week, and Alibaba released Qwen 3.8 this Monday. Silicon Valley and Washington started talking about the models immediately, especially K3, which is widely seen as the best of the bunch. David Sacks, a venture capitalist and AI adviser to President Donald Trump, called the performance of Moonshot's model "concerning." Earlier this week, Commerce Secretary Scott Bessent suggested the US might impose sanctions on Chinese AI companies. On Wednesday, Michael Kratsios, director of the White House Office of Science and Technology Policy, alleged that the Trump administration has "information that Moonshot AI distilled Anthropic's Fable for the development of its K3 model," which he said amounted to "stealing proprietary US technology and undermining American research" and was "unacceptable." (Moonshot AI did not immediately respond to a request for comment.) The new Chinese models have a few things in common: Third-party benchmarks show that they perform nearly as well as the best Western models; they are optimized for agentic coding tasks (the hottest thing in AI this year); and they are or will soon be released with open weights, making them accessible and transparent. But perhaps the biggest parallel between the current moment and January 2025 -- when the world was shocked by DeepSeek's R1 model -- is that it reaffirms how American and Chinese AI labs are taking diverging paths when it comes to being open or closed. When it first burst onto the scene, DeepSeek challenged the premise that only closed-source models built with billions of dollars of investment in compute infrastructure and training could achieve frontier performance. But since then, Western AI labs have continued developing AI the same way, and now American frontier models feel more roped-off than they were a year ago. Anthropic said for months that its latest Mythos model was so dangerously good at hacking that only approved collaborators could use it. When it was finally released more widely, the White House responded by issuing broad export controls, which forced Anthropic to take Mythos and its less capable sister model, Fable 5, offline temporarily. OpenAI similarly delayed the release of GPT 5.6 after it received a request from the White House. In China, meanwhile, the situation looks very different. Chinese startups and tech giants have doubled down on open source: Anyone with a good enough computer environment can now download an open-weight model, run it locally, add customizations, and overall enjoy a much greater degree of freedom than OpenAI and Anthropic would ever allow. In many ways, the open versus closed debate is more entangled with the US versus China debate than ever before. There are a lot of reasons why Chinese labs have chosen a business strategy built atop open-source models. Being the newer, smaller fish in the AI field, making their models free and open can help Chinese firms attract more users, collaborators, and media spotlight. It also puts them in a separate lane of competition from the one that OpenAI, Anthropic, Google, SpaceX, and other deep-pocketed giants are in. Earlier this year, rumors spread that Alibaba might be considering joining the closed-source race after it rearranged its corporate AI model development teams. But the tech giant announced on Monday that it would again release the latest version of Qwen -- its line of open-source models beloved by the global tech community -- with open weights, signaling to customers and the public it is not pivoting away yet. Perhaps the strongest validation of the Chinese approach is the models themselves. Chinese labs have produced what are widely considered the world's best open-source AI models, showing that American companies no longer have an exclusive hold on building the most capable systems. Arena AI, a crowdsourced model evaluation platform, now ranks K3 as the best model when it comes to web development tasks and number four in agentic tasks, just below Anthropic's Fable and Opus 4.8, as well as OpenAI's GPT 5.6. Artificial Analysis, an independent AI benchmarking company, ranks K3 in the third spot in its intelligence index. Shortly after Moonshot AI released a preview version of K3 on July 16, people around the world began rushing to try it out, consuming so much inference computing resources that the company is temporarily restricting new users from signing up. As more people start using open-source Chinese models with capabilities nearly on par with that of their Western competitors, some have started questioning whether paying for OpenAI or Anthropic's offerings is really worth it. The fact that multiple Chinese labs were able to create capable agentic models and were unafraid to release them to the public is poking holes in the popular understanding that OpenAI and Anthropic are miles ahead of the competition. Just like with DeepSeek, K3 fans are now celebrating it as evidence that Western AI models are overhyped and overprotected. "It is absolutely wild how much love Kimi got," Rui Ma, founder of the independent research firm Tech Buzz China, said in a social media post, referring to the company's announcement that demand for K3 had overwhelmed its servers. "Only made possible by the poor comms and decisions from [Silicon Valley] labs in the past year," she added. "I think Anthropic has overhyped the risks, or described risks that are coming soon but do not currently proliferate," says Nathan Lambert, an independent AI researcher in Seattle who recently visited Moonshot AI's office in China. He admits, though, that just like the rest of the general public, he has little firsthand information about how Mythos really performs. "We rely on a few private companies and a federal government with depleted state capacity to make that judgment call," he says. In addition to challenging the mainstream Western narrative about AI, Chinese open-source models are also proving to be a practical commercial replacement for American closed models. They aren't just talked about on X and benchmarked against other models -- they have become a popular and readily usable option for a growing number of Western startups and individual users. "There's a real shift toward actually using these newest models that started with GLM 5.2," Lambert says, referring to Z.ai's model. "Even weeks after the GLM 5.2 release, I hear from AI researchers in the Bay Area that they are still using it for a core part of their workflow. Kimi, being a stronger model, will only do more, especially in areas like cybersecurity, where Mythos, Fable, and GPT 5.6 are effectively unusable." In fact, this is already happening. On Tuesday, OpenAI disclosed a concerning incident in which its GPT-5.6 Sol model hacked into the production system of Hugging Face, an open-source AI platform. Hugging Face has said it resorted to using the open-source model GLM 5.2 to analyze the cyberattack because other frontier models were refusing to help due to built-in safety guardrails. Chinese AI models also tend to be cheaper than Western alternatives, but that's not necessarily their main selling point. While models like K3 charge less for tokens, early tests show that they may require more tokens than Western models to solve the same problems, thus making the cost gap smaller. "In my fairly limited use, [K3] also seemed very token-hungry. It's not obvious to me that this model is actually that cheap to run," wrote Dean Ball, a former White House AI adviser who recently joined OpenAI as its head of strategic futures, in a social media post in which he also praised K3 as "a very good model." Even if they are relatively token-hungry, models like K3 do ultimately challenge a fundamental assumption that has driven OpenAI's and Anthropic's strategy for years: that AI labs need infinite funding to scale up their compute capabilities just to make better models. "In the end, open-weight models deter further AI capex," Ball wrote.

Kimi: Threat or menace?

Chinese company Moonshot AI released a new version of its Kimi model this week, generating another wave of discourse about China and open source AI. Moonshot said that although Kimi K3 "still trails the most powerful proprietary models, Claude Fable 5 and GPT 5.6 Sol," the new open source model "demonstrated frontier-level performance across our evaluation suite, consistently outperforming other tested models." Independent analyses from Arena.ai and Vals AI also suggested that Kimi is competitive with flagship frontier models. The announcement, which coincided with a speech from Chinese president Xi Jinping at the World AI Conference in Shanghai, seems to have spooked Wall Street, with the Nasdaq dropping about 1% on Friday as investors sold off stocks in chip companies like Nvidia. Many of the resulting posts from tech industry figures will sound familiar to those who remember the debate after another Chinese company, DeepSeek, released its open source R1 model in January 2025. Except now, everything seems heightened after the Trump administration's tariff war with China, repeated fights over the national security threat supposedly posed by Anthropic, and as major AI companies prepare to finally go public. For example, David Sacks -- the Trump administration's former AI czar and now co-chair of the President's Council of Advisors on Science and Technology -- contrasted Kimi's progress with a United States that is "tying itself in knots: politicians and bureaucrats are banning new data centers, piling on state regulations, and pushing for new federal agencies to pre-approve frontier models. This is how you lose the AI race." (The news also gave him an excuse to take a dig at Anthropic, calling Claude an example of "woke lobotomized models.") And former Uber CEO Travis Kalanick echoed complaints that Chinese are "distilling off" (i.e., being trained on the outputs of) American AI models. "If distillation isn't enforced against, then everyone should be able to distill from everyone else.. otherwise one arm [would be] tied behind American models' backs," Kalanick wrote. (Of course, American models have also been built on top of Chinese ones, specifically Kimi.) Meanwhile, OpenAI's head of strategic futures Dean Ball said that Kimi is "a very good model" whose performance probably can't be "explained away by distillation or anything like that," adding that he's "personally surprised the Chinese state continues to allow the open sourcing of models this good, given potential risks." In fact, Ball suggested that "probable outcome of an open-weight-model-dominant world is full AI communism," where AI is treated as "a 'public good' which will ultimately be provided by the state as a kind of 'digital public infrastructure.'" "This future strikes me as a dystopian hellscape, but I've never met an open-weight models advocate who doesn't ultimately concede this is where things end," said Ball. He even suggested that the Trump administration (which he used to work for) will eventually realize it needs to "create large amounts of regulatory risk around the use of open-weight Chinese models." "You don't need to 'ban open source' (one of the dumber motifs of AI policy discussion)," Ball said. "You just need to direct every agency to issue soft law that creates FUD [fear, uncertainty, and doubt]. 'A Federal Reserve Advisory Bulletin found that there may be backdoors in Chinese AI models.' It needn't be that well justified. You just create enough regulatory risk that every regulated enterprise backs off." However, Shakeel Hashim, editor of the AI-focused publication Transformer, argued that much of the worry is overblown, both because Kimi "likely does not have dangerous cyber capabilities," and because the Chinese government will face "extremely similar incentives" to restrict open Chinese models once they develop those capabilities.

America needs to stop getting shocked by Chinese AI

Last week, two Chinese AI companies unveiled models they say can credibly compete with the best systems from OpenAI and Anthropic. The response was swift and predictable. Markets wobbled, commentators declared Silicon Valley shooketh, and policymakers reached for the familiar language of arms races and wake-up calls. In one headline, The Associated Press said a Chinese model had taken the "US tech industry by surprise." Bloomberg described it as a "surprise breakthrough" that is "roiling markets" and sending global tech stocks tumbling over concerns it could force US firms to rethink their gargantuan spending on data centers, chips, and other AI infrastructure. Business Insider questioned whether the launch is "The next DeepSeek?", referring to the Chinese model that blindsided the US AI industry last year. Xprize founder Peter Diamandis went as far to call the release America's "AI Sputnik moment," referring to the Soviet satellite launch at the height of the Cold War that encouraged significant US investment in its science and space programs. Of course, DeepSeek was also widely described as America's AI Sputnik moment, a comparison that felt less gratuitous then as DeepSeek appeared to arrive with little warning, challenged the prevailing assumptions about the costs of frontier AI, and prompted immediate reactions across the technology and financial sectors. What is actually surprising is that the model announcements were a surprise at all. For years, we have been warned that China was catching up in AI. Yet the world is shocked when it starts to look like the moment may have arrived. US and Chinese companies train almost all of the world's most-used AI models, and six of the top 10 AI tools on OpenRouter's leaderboard tracking token consumption and benchmarks were Chinese. The performance gap has been narrowing for some time, with recent models from companies like Z.ai and DeepSeek seen as highly competitive with top-tier offerings from US labs like Anthropic and OpenAI. Chinese models are also significantly cheaper to use, and reports suggest US companies are increasingly turning to Chinese tools as the cost of using domestic providers surge. Beijing has also been keen to support homegrown AI efforts, including incentivizing and funding innovation and cracking down on firms trying to shed their ties to China. Meanwhile, Washington's AI strategy has often veered between heavy-handed intervention that has left allies questioning America's reliability and a laissez-faire assumption that markets will see things right. It is a difficult approach to maintain against a competitor prepared to mobilize the full force of the state behind a single technological goal. Beijing-based startup Moonshot AI, one of China's leading AI model developers, unveiled a new flagship model on Friday, claiming it outperforms nearly every US model, trailing only OpenAI's GPT-5.6 Sol and Anthropic's Claude Fable 5. Moonshot is also pricing Kimi K3 aggressively, charging $15 per million output tokens, compared with roughly $30 for GPT-5.6 Sol and $50 for Fable 5. Demand was so strong after the launch that Moonshot, the company claimed, that it temporarily paused new subscriptions after the service was overwhelmed. The majority of responses mainly focus on this release. Days later, Chinese tech titan Alibaba followed with a preview of Qwen3.8. It described the new model as "one of the most powerful model[s] available today" and "second only to Fable 5." This only added to the uproar Kimi K3 had caused. Crucially, both companies plan to make their new flagship models publicly available. Both Moonshot and Alibaba say they plan to release their models as open weight, which would allow developers to download, use, and modify the core values created during the AI's training that shape its responses. It stands in stark contrast to the closed, proprietary approach to frontier models taken by most leading US AI labs, including OpenAI, Anthropic, and Google. The economics deserve particularly close scrutiny. There's the whole unsettled debate over whether, and to what degree, Chinese companies are -- as American firms accuse -- using US models to train their own, which could improve performance at a fraction of the cost. Tokens are not directly comparable between models, and token prices alone give an incomplete picture of how much it costs to use an AI system. A more expensive model may, for example, generate better responses with fewer tokens. Companies also routinely subsidize inference costs to win over customers. Cheaper, in other words, does not automatically mean better, or even less expensive overall. Still, the possibility remains that Chinese labs may eventually produce models that are not merely cheap substitutes, but systems that could genuinely match or outperform their US rivals. Even companies that trail the frontier slightly could still have an enormous impact if their models are good enough, easier or cheaper to deploy, or available on more attractive terms. This could have direct consequences for US companies, the wider economy, and national security. Anthropic and OpenAI are both gearing up for what could potentially be trillion dollar IPOs, valuations that in part depend on the expectation that they will dominate the global AI market. Capable Chinese models challenge that assumption, and could potentially draw away customers, squeeze margins, and weaken growth assumptions underpinning those valuations. Given how expensive American AI has become, some US startups are already reportedly turning to cheaper Chinese models. There is a wider market risk, too, reaching far beyond a handful of AI players. Tech stocks make up an outsized share of US markets, and much of that recent growth has been tied to expectations that AI demand will continue to soar. Companies have piled hundreds of billions of dollars into data centers, chips, energy, and other infrastructure that relies on the assumption American firms will continue to dominate. If Chinese labs can capture some of that demand, or show models that can be produced and operated for less, investors would inevitably question whether those costs are justified. Given the money involved, any reassessment on their part would have ripple effects throughout all of these industries, as well as the millions of people with savings or pensions exposed to them. There are security considerations, too. Highly capable open Chinese models, even if trailing the US frontier, could make advanced AI systems available to a much wider range of users, notably in cases where US companies restrict access or impose stronger safeguards. When the US government demanded Anthropic limit access to its latest models, cybersecurity leaders warned that doing so would make it harder for defenders to find and fix vulnerabilities. Those restrictions are harder to justify if comparable models are available elsewhere. Organizations denied access to US models may feel compelled to rely on Chinese alternatives to secure their networks, or else accept the greater exposure to attackers able to use the same tools. Already, reports are starting to emerge where Kimi K3 identified and fixed cyber vulnerabilities that OpenAI's Codex and Anthropic's Fable would not touch due to safety guardrails. Even less broadly capable models can still pose a threat and some already appear to be doing so. In June, China's Z.ai claimed its GLM-5.2 model could match Anthropic's Mythos on cybersecurity tasks, even though it trailed in more general tasks. As neither model has yet been fully released, it is still difficult to independently assess how capable either actually is, and companies' benchmark claims should be treated with caution. Even so, there has been little public suggestion that the companies are fundamentally misrepresenting their results when it comes to performance. But the exact ranking is almost beside the point. Whether Kimi K3 and Qwen3.8 ultimately prove to rank among the world's top five models or merely the top 10, the broader conclusion remains the same: China's leading AI companies are now producing systems that could plausibly rival those emerging from top US labs. And they are doing so with enough regularity that each new release should no longer be treated as a shock, let alone something as singularly galvanizing as another "DeepSeek" or "Sputnik moment." If this really is a race, it's time to accept that someone else might actually win, or at least get close enough that they might as well have.

Moonshot's upcoming Kimi 3 is expected to close the gap with Anthropic's Opus 4.8

The latest iteration of Chinese AI lab Moonshot AI's Kimi model series is expected to perform at par with or even surpass Anthropic's Opus 4.8, the Financial Times reported, citing anonymous sources. Moonshot's Kimi K2 models have been received well in the open-source AI market, ranking high on benchmarks and demonstrating capabilities that aren't too far behind the latest frontier models. The company's upcoming release, called Kimi K3, is said to take this one step further to close the gap with closed-source models from the likes of OpenAI and Anthropic. The FT reports Kimi K3 will be the largest open-weight AI model from China, with a parameter count between 2 trillion and 3 trillion, and will be released "in the coming days." Moonshot is also said to be raising fresh capital in a round that would valuate it at $31.5 billion. The company in May raised $2 billion at a $20 billion valuation. The news comes amid a fresh debate on the value of paying AI labs like OpenAI and Anthropic for their expensive, closed-source models. Industry leaders fear that AI labs will somehow manage to extract the data their clients submit for use with their AI products like ChatGPT and Claude. Executives are pitching their own products as alternatives, or recommending companies to take cheaper open source models, like those developed by DeepSeek, Z.ai or Moonshot, and train it for their own purposes. The argument has gained momentum, especially as open models from China close the gap with their more expensive, frontier counterparts.

Jensen Huang argues American companies should be allowed to use Chinese AI models -- Nvidia CEO says backdoors connected to China are misconceptions

He also believes that American AI should be accessible to everyone. Nvidia CEO Jensen Huang thinks that American companies should be allowed to use Chinese AI models, even as Washington is trying to ban them. When Axios co-founder Mike Allen asked Huang in an interview if Americans companies should be allowed to use Chinese AI models, Huang responded with "absolutely." The answer comes right after Chinese firm Moonshot AI released a 2.8T open-weight model called Kimi K3, which -- although it isn't as powerful as frontier models like Fable 5 -- is comparable to GPT 5.5 and Claude Opus 4.8 while costing just a third of these models. One of the biggest concerns of U.S. leaders have is that these AI models might come with vulnerabilities that the Chinese government can use to attack American interests, but Huang said that this is an incorrect assumption. "There is a misconception that somehow there are backdoors that are somehow connected to China in some way," said the Nvidia chief. "You download the models, you can fine-tune it, you can enhance it, you can guardrail it as you desire." Huang shares the same sentiments about American AI models. Just last month, the U.S. enforced an export restriction on Anthropic's Mythos and Fable 5, citing security threats -- although access was eventually restored after its developer placed a filter to block these tools from identifying software vulnerabilities. OpenAI's ChatGPT-5.6 received the same treatment, and Washington warned the firm that it should not release its latest model without getting the green light from the government. Huant argues that, instead of restricting access to these powerful models at launch, AI firms should make their models available to all and make them more secure through rapid testing and fixes. But even as he advocated the need for everyone to have access to closed models, Jensen also noted that various industries, such as the sciences and cybersecurity, need open models as well. He claims that these models make AI more secure, as other people can inspect them to look for weaknesses and fix them as required. "If everything just becomes one single model, one single point of attack, one single source of failure, I think the world is much, much more vulnerable," Huang said. As for the market's negative reaction every time cheaper, open-weight models become available, the Nvidia CEO says that investors misunderstand their impact. Huang said that this happened when DeepSeek arrived for the first time, and it's happening again with the arrival of Kimi. He says that these open models, which cost less to run, will encourage more people to use AI. So instead of cutting data center demand, these cheaper, more efficient models are actually good for the industry in general because they will drive demand. (And with higher demand, there's more incentive to build data centers and buy AI GPUs, which is ultimately good for Nvidia.) Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.

The truth nobody wants to admit: Chinese or not, open models are competitive now

OPINION Every six months or so a Chinese model sparks a panic, calling into question America's AI dominance. Moonshot AI's Kimi K3 is the latest example. Recall when DeepSeek R1 shook markets early last year? Following a similar pattern, Moonshot's latest model isn't all that interesting apart from its benchmark performance, and at nearly three trillion parameters it's beyond the reach of most enterprises. For the likes of OpenAI and Anthropic, the technology isn't as interesting as its national origin. A Chinese model can be dangled as a threat to national security -- Anthropic's fearmonger and chief Dario Amodei has said as much in the past. Last month the executive accused Alibaba, another prolific Chinese AI model dev, of using Claude to improve its Qwen family of models through a process called distillation. Those claims are rather rich coming from a man whose company just agreed to pay to settle claims over vacuuming up millions of pirated books to train Claude. Pot, kettle black much? With that said, the anxiety caused by Moonshot's Kimi K3 does feel different. The model's size is a factor. At 2.8 trillion parameters, Kimi K3 is the largest open weights model ever built. The model's girth certainly reinforces the idea that K3 is a frontier model. There are interesting architectural changes to the model, but from what we can tell, they aren't at all related to the anti-China-AI rhetoric making the rounds right now. But let's talk about the benchmarks! We've been down this road before with Z.AI's GLM family, MiniMax M-Series, Qwen, and of course DeepSeek. A new model is released alongside benchmarks that show it trading blows with OpenAI, Anthropic, or Google's best. Kimi K3 is following that well-trodden path. Last week, Moonshot published a lengthy blog post packed with benchmark charts and demos showing off its new model, and predictably, told the same story as always: Despite US trade restrictions on American-made accelerators, Chinese model houses aren't as far behind as Altman and Amodei would like. What has changed is Uncle Sam's attitude toward frontier models. GPT-5.6 was apparently concerning enough that the US government delayed its release, while Claude Fable 5 was pulled offline shortly after launch as officials investigated security concerns. As we later reported, the boogeyman hiding under Fable's bed wasn't some Lovecraftian horror. It didn't even rise to the level of a Disney villain. According to one researcher, it was the sudden realization that generative AI models have achieved a level of competency that users can type "fix this code" and they'll do just that. Yet, the disruption, however brief, may as well have been free marketing for Anthropic, which has been peddling the idea that open models are inherently dangerous. In this respect, the drama around Fable 5 and GPT-5.6's delay has become ammunition. If America's top models are enough to worry the US government, why shouldn't Kimi K3? After all, OpenAI and Anthropic are American companies that can be bent to the Trump administration's whim. But an open weights frontier model that originated in China? Not so much. The Trump administration has apparently taken notice. White House officials are reportedly hearing calls to restrict access to Chinese models, but haven't taken formal action yet. Regardless of what happens this time around, new reports suggest the US and China will meet later this year to discuss the growing threat of their respective nations' AI endeavors. In an AI arms race, an AI summit was inevitable. The US model houses would certainly benefit from restrictions on Chinese models. Under the guise of national security, the restrictions would cut off Chinese AI developers from becoming the US labs' biggest competitors. But less competition inevitably means enterprises and consumers get screwed. The US hasn't pursued large open weights models on the same scale as Chinese devs. Thinking Machines Lab's newly announced Inkling model at just shy of a trillion parameters is about as good as it gets for American open-weights models. The next closest would be Nvidia's Nemotron 3 Ultra at 550 billion parameters. Who knows whether the White House will actually fall for the ploy and give Amodei the knee-jerk response he presumably is looking for. But if the Trump administration doesn't want US companies, particularly those serving government agencies, using Chinese models, the answer isn't less competition. It's more. ®

Chinese AI start-up Moonshot to launch model challenging Anthropic's lead

Chinese AI start-up Moonshot is set to release a large language model with capabilities approaching those of frontier US labs such as Anthropic, as the gap narrows between the two countries on state-of-the-art AI. The Beijing-based group is scheduled to release Kimi K3, China's largest AI model to date with 2tn-3tn parameters, in the coming days, according to two people familiar with the matter. The number of parameters refers to the size of the model's neural network, with a higher count generally leading to greater capabilities. Anthropic has not disclosed its models' parameters, but industry experts speculate Opus 4.8 has 1.5tn-2tn parameters. K3 is expected to exceed the performance of Opus 4.8 by mainstream benchmarks, said the people, who added that it would be freely available to download as a so-called open-weight model. However, its capabilities are likely to fall short of Fable, a powerful model that Anthropic has suspended after the US raised concerns over its hacking capabilities. The launch of K3 could challenge the industry consensus that Chinese AI models are eight to 12 months behind US ones in terms of performance. Being open weight would also pose a significant challenge to US labs such as Anthropic and OpenAI whose expensive frontier models remain closed. While the latest models from the US continue to outperform Chinese tools at the most complex tasks, a growing cohort of US tech investors and executives has warned that the gap is narrowing. Marc Andreessen, co-founder of US venture capital group Andreessen Horowitz, wrote last month that GLM-5.2, released by Chinese lab Z.ai, was "the first Chinese AI model to match and often beat the American big lab public AI models with no compromises". Companies from Silicon Valley to Europe are also switching to cheaper Chinese models to reduce their rising bills for technology from US labs. The US's top model makers have poured hundreds of billions of dollars into building out infrastructure and developing the most advanced AI tools. They have begun to charge steeper fees for companies to access them this year. Anthropic will increase the price of Opus 4.8 by 50 per cent to $3 per million input tokens and $15 per million output tokens in September, according to its website. Chinese labs such as Moonshot and DeepSeek, meanwhile, have released open-weight models that are cheaper to run and can be downloaded and modified by users. Moonshot's K2.6 model, for instance, costs about a third of Anthropic's Opus 4.8. Anthropic in February accused Chinese AI labs of conducting "industrial-scale distillation attacks" on its models. Distillation refers to the practice of training smaller models on the outputs of more advanced systems, allowing developers to replicate high-level performance without the same computing resources. Some Chinese companies have pushed back such claims, calling it an excuse to protect their monopoly. The valuation of Chinese AI labs has increased significantly this year but remains a tiny fraction of their US peers. Moonshot is raising a new round of funds that would value the company at about $31.5bn, one of the people said. Meanwhile, DeepSeek is starting a new round at about $71bn, the FT reported this week. Anthropic reached a valuation of $965bn in its recent fundraise in May, while OpenAI's latest valuation was $852bn.

Kimi K3 rocks the AI industry as Moonshot AI undercuts closed-source American competitors on price

Memory makers are the big winners, regardless of how it works out. A new AI model from Chinese firm Moonshot AI has had its "DeepSeek moment," causing major disruption in the global AI market and spooking Western developers. Kimi K3 is an open-weight model, with 2.8 trillion parameters, making it the largest open-weight AI model ever released. Internal benchmarks have it competing with models like GPT 5.5 and Claude Opus 4.8, and Arena.ai awarded it the number one spot in its Frontend Code Arena test, even beating out Claude Fable 5. It doesn't win every benchmark, and reports that suggest Kimi K3 is much slower than frontier models from companies like Anthropic and OpenAI. All benchmarked results are drawn from API access, too, so can't be verified until Moonshot releases the weights on July 27. But that hasn't reduced the impact of this model's release on the AI industry. With Kimi K3 cutting costs compared to the competition, it's drawing a lot of interest from companies hoping to reduce AI spend. For comparison's sake, OpenRouter tables Kimi K3 at $3/15 per million inputs and outputs. OpenAI's GPT 5.6 Sol is more expensive than that, at $5/30, and Anthropic's Claude Fable 5 is $10/50. So, it's fair to say that Kimi K3 is incredibly competitive on price, especially when tabled against the costs of those closed-source Western AI models. Microsoft is also considering Kimi K3 for Copilot, while the White House may ban Chinese models entirely. Meanwhile, memory makers are rubbing their hands together with glee, as Kimi K3 occupies up to 1.4 TB of memory, given its huge number of parameters. Fast, cheap, or American? The past few months have been full of talk about the frontier AI models from Anthropic and OpenAI. Mythos was big and scary until OpenAI had something equivalent. Then Fable debuted, and it was even better but not so scary anymore. Apparently. But these models were also proving very expensive to run, at a time when companies with big AI deployments were questioning the return on that investment. Uber limited AI use by developers, and others killed the AI-boosting leaderboards they'd championed towards the end of 2025. So when Moonshot debuted Kimi K3 with running costs a third that of western frontier models for the same results, the world took notice. In much the same way as DeepSeek's R1 debut in 2025 showcased how models could be trained for less -- even if there may have been some corporate espionage involved -- and Kimi K3 is holding up a similar mirror to Western frontier developers. Where DeepSeek R1 was lean, though, Kimi K3 is huge -- so large, the developers are calling it the first open 3T-class system, and China's largest AI model to date. According to Bloomberg's sources, its sparsity ratio is the highest yet seen by any AI model. That's the measurement of how many parameters are activated for each task relative to the model's size, showcasing both Kimi K3's overall size and its impressive efficiency in the same breath. This doesn't eclipse the most capable models from companies like Anthropic and OpenAI in every test. Arena.ai's rankings put it within the top 10 on most of its tests, but only coming out on top in a couple. But if Kimi K3 can offer results comparable to more expensive alternatives like Claude and ChatGPT, it's likely to draw a lot of interest from Western companies, and it appears to be already doing so. Enough that it's revived calls for the U.S. to gatekeep access to international AI models, in a similar manner to how it recently pushed for companies to share exclusive model access with the U.S. government before a wider release. In comparison, Moonshot is opening up Kimi K3 to the wider world. As part of releasing the model weights to the public, it will allow companies and organizations to run the model themselves without using Moonshot's cloud services, making adoption easier and potentially cheaper. But it won't be cheap, as Kimi K3 still needs serious hardware investment to get up and running, by virtue of its massive VRAM requirements alone. A Win for (Chinese) Memory Makers As large companies with major AI deployments began to scale back their AI initiatives in 2026, there's been a growing concern that all that infrastructure everyone's been spending hundreds of billions of dollars on might not be needed. Meta just started selling excess compute in a pivot to cloud services, and xAI unloaded the entire compute capacity of Colossus 1 to Anthropic at a discounted rate. But if Kimi K3 is the way the industry might go, hardware demands are unlikely to fall, and as Jevon's paradox suggests, greater efficiency is only likely to increase usage, not shrink it. Those trillions of parameters need to be stored in memory, and Bloomberg's estimates suggest Kimi K3 will require close to 1.5 TB of memory. It would need masses of high-end Nvidia GPUs to deploy it effectively, making the number of companies and organizations that could actually run Kimi K3 at scale rather small. So even those who do look to leverage Kimi K3 to reduce operating costs will still need powerful hardware, and specifically a lot of memory. This suggests that the major competition for cutting-edge models is not going to crater costs like we initially saw with DeepSeek R1 last year, which means memory makers are going to continue making money hand over fist, due to their outsized demand and limited supply. But Chinese memory suppliers like CXMT are on the rise, and on track to eclipse Micron's DRAM wafer capacity by the end of the year. Smaller local AI models will also continue to be further optimized for domestic hardware, reducing the stranglehold that some large tech companies have on the AI supply chain. Competitive, efficient, but unwieldy Kimi K3 is an industry disruptor and is already raising questions over AI costs, capabilities, and access. It's shown that you don't need proprietary models locked to a specific service to achieve frontier-model capabilities. It's also cheaper to run, but Moonshot achieved this with a sparse model that still requires massive hardware investment to operate. Even though Kimi K3 activates only a fraction of its trillions of parameters for each query, it still needs all of them to be stored. Deploying this model at scale requires substantial memory capacity, bandwidth, and interconnects, even if its compute demands aren't as strenuous. The open-weight nature means it has very real potential to supplant usage away from Western frontier models in the short term, but it isn't about to change the story we've been told on required infrastructure. Kimi K3 needs the same kind of hardware to run as GPT 5.6 and Fable -- which is likely to be far more of a limiting factor on its adoption than any kind of government blocks.

Chinese AI model takes US tech industry by surprise with abilities rivaling Claude and ChatGPT

Another powerful new artificial intelligence model from China took the U.S. tech industry by surprise Friday, the latest sign that Chinese startups that publicly release their "open-source" AI technology are making the California titans of AI sweat. The newest Kimi K3 model from Beijing-based startup Moonshot, run by a Pink Floyd-loving entrepreneur who earned his doctorate in Pittsburgh, appears to be catching up to the best versions of Anthropic's Claude and OpenAI's ChatGPT. "This may be the single biggest release of the year," and marks a moment when open-source Chinese models are surpassing closed U.S. models, said Anastasios Angelopoulos, co-founder and CEO of Arena, a platform for evaluating AI systems. Kimi K3 topped the charts on Arena's ranking of what it calls "front-end coding capability," which is one measure of an AI large language model's performance. "More results are rolling in that are likely to continue to show it is at the top of the pack," Angelopoulos said on social media. It was not likely a coincidence that K3's unveiling came shortly before Chinese President Xi Jinping's opening address Friday to the nation's annual World Artificial Intelligence Conference in Shanghai. American-led restrictions have blocked China from accessing some of the world's most advanced technologies, spurring China's efforts to build its own know-how and intensifying the rivalry between the world's two biggest economies. "The development of artificial intelligence should not be a solo performance by any single country but rather a symphony of global cooperation," Xi said at the event. Chinese AI models have shown large strides K3 follows another major AI model release last month from the Chinese startup Zhipu, or Z.ai. Its new flagship GLM-5.2 model is already being widely used by software developers around the world who say it can perform work almost as good as the top U.S. models at a cheaper price. The hype over the new Chinese model resembles the market-shaking panic that followed the release of a new model from Chinese startup DeepSeek in early 2025, though not everyone finds it justified. The response to K3 is an "overreaction shockingly similar" to DeepSeek's release last year, said tech analyst Patrick Moorhead on social media. He said it could be good for parts of the broader AI industry but pose a revenue challenge to Anthropic and OpenAI. During the conference that runs until Monday, tech giant Huawei has also been showcasing a new AI computing system called the Atlas 950 SuperPoD, a signal that China increasingly is amassing the domestic hardware it needs despite U.S. restrictions on imports from chipmakers like Nvidia. Moonshot hasn't said what hardware it used to build K3, but the startup is a partner with Huawei. The price to use K3 is highest yet for a Chinese AI model, but it is still half as expensive as OpenAI's high-performing GPT-5.6 Sol model, according to a Friday report by Bank of America research analysts. U.S. politicians and several major U.S. AI companies including Anthropic and OpenAI have accused Chinese AI models of illicit "distillation" of their models to extract their technologies, a claim that Beijing says is "groundless." Anthropic in February accused DeepSeek, Moonshot and a third China-based AI lab, MiniMax, of engaging in campaigns to "illicitly extract Claude's capabilities to improve their own models" using the distillation technique that "involves training a less capable model on the outputs of a stronger one." Anthropic said that distillation can be a legitimate way to train AI systems but it's a problem when competitors "use it to acquire powerful capabilities from other labs in a fraction of the time, and at a fraction of the cost, that it would take to develop them independently." But it can go both ways. San Francisco-based startup Anysphere, maker of the popular coding tool Cursor, has acknowledged that one of its top products was based on Moonshot's K2.5 model. Elon Musk's SpaceX is planning to close a deal to buy Cursor for $60 billion later this year. K3 marks a leap for 'open-source' AI models Moonshot co-founder and CEO Yang Zhilin earned his Ph.D. in 2019 at Carnegie Mellon University, where he is said to have made fundamental contributions to the machine-learning field and was known for a love of rock bands like Pink Floyd. The pride among his former colleagues at the school in Pennsylvania transcends the U.S.-China rivalry. "What a huge win for the open-source community! It feels like just yesterday Zhilin was graduating from my lab at CMU," wrote his former adviser Russ Salakhutdinov, who is also a former director of AI research at Apple. Developers who build "open-source" AI make key components of the technology accessible for anyone to examine, modify and build upon. Proponents say open-source practices promote innovation, while critics warn that making powerful AI models publicly accessible poses safety and security dangers. ___ Associated Press writer Chan Ho-Him contributed to this report.

Chinese startup Moonshot AI unveils Kimi model it says rivals OpenAI, Anthropic

Chinese startup Moonshot AI has unveiled a new model it says closes the gap with leading U.S. offerings and surpasses OpenAI and Anthropic's most capable systems on some benchmarks. Kimi K3 still trails Anthropic's Claude Fable 5 and OpenAI's GPT 5.6 Sol on overall performance, the company said on Friday, but consistently outperformed other tested models. The model beat Claude Opus 4.8 and GPT 5.5 -- models that sit just behind Anthropic and OpenAI's leading-edge systems -- on benchmarks including coding and general agents, according to Moonshot. It's China's largest AI model so far, with 2.8 trillion parameters, referring to the size of its neural network. "Despite persistent hardware/compute capacity constraints in China, K3 demonstrates that pre-training scaling, paired with architectural innovation, can still deliver step-change gains for flagship Chinese models," Bank of America analysts said in a note led by Alex Liu. The release comes as the race for AI supremacy between the U.S. and China intensifies. Chinese AI models are already gaining traction among Western companies as they close the performance gap with U.S. rivals and remain cheaper to use than the most advanced offerings from American labs. U.S. lawmakers are considering how to curb the growing adoption of Chinese AI models by homegrown companies.

Nvidia's Jensen Huang defends Chinese AI: "Open-source models that are excellent should be used"

Serving tech enthusiasts for over 25 years. TechSpot means tech analysis and advice you can trust. Big quote: Jensen Huang just spent a week watching his company's valuation get hit by the threat of a Chinese open-weight model. His response was to go on the record defending it. Speaking to Axios in Fort Worth, Texas, at the opening of a new phase of the Wistron plant that builds Nvidia's AI infrastructure, the Nvidia CEO said American companies should "absolutely" be free to run Chinese models. "These Chinese models are excellent," he said. "Open-source models that are excellent should be used." Asked whether China could displace American labs, he was blunt: "Zero possibility." The timing is what makes it interesting. Moonshot AI released Kimi K3 on July 16. Independent evaluators put it third on Artificial Analysis' Intelligence Index, only behind Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 Sol, and first on LMArena's blind Frontend Code Arena board. The market took it about as well as it took DeepSeek. The Philadelphia Semiconductor Index fell 12.5% in a week, its worst stretch in 15 months. Taiwan's benchmark dropped more than 6%, Japan closed down 4%, and Chinese AI stocks actually got hit harder than American ones, with Zhipu down as much as 30% in Hong Kong and MiniMax off 16%. "The market misunderstood the impact of DeepSeek the first time," Huang told Axios, adding that Wall Street has "misunderstood the impact of Kimi again this time." His pitch is one Nvidia has made before, delivered without much subtlety this time: "Free AI should be great for hardware. Free AI should be great for chips. Free AI should be great for data centers." The logic is that as cheap, capable models get used more, not less, more usage means more chips running somewhere. He also argued open models don't cannibalize OpenAI and Anthropic, but they give people a free first taste, and plenty of those people end up paying for something faster and more reliable. Bloomberg's analysis of K3 noted that where DeepSeek's story was about cheaper training, Moonshot's is about a much larger model that leans harder on memory infrastructure. The security argument, flipped Huang waved off the idea that a downloaded Chinese model is some kind of backdoor to Beijing, pointing out you can inspect the weights, tweak them, and run the whole thing sealed off from the internet if you want. His argument is that openness actually makes things safer, because more people are looking for problems. Lock everything into one closed system, he said, and "if everything just becomes one single model, one single point of attack, one single source of failure, I think the world is much, much more vulnerable." Then he turned that same logic on an American company. Huang said Anthropic should open up Claude Mythos, its cybersecurity model that's currently restricted to a vetted group of partners, calling it something that "should be available as a service" and arguing "holding Anthropic back is not in the benefit of the United States." His framing: "Just because Mythos is not available, open models are available anyhow." It is worth noting that Nvidia is one of the partners with access to Mythos already, through Anthropic's Project Glasswing program. Glasswing has grown from around 50 partners to roughly 200 across about 15 countries, and Anthropic has said models this capable will likely be widely available within 6 to 12 months regardless of what it decides. Meanwhile, in Washington... Huang's comments landed hours after Treasury Secretary Scott Bessent told Fox Business the administration is looking into whether Chinese AI models were built on stolen US intellectual property. "If we see ... that overseas models are stealing from our great companies, we have the ability to sanction them because of this theft," Bessent said, pointing to what he called "watermarks" of US models showing up inside Chinese ones, with action possible within days or weeks. He also floated whether US companies should have to disclose to customers when they're running Chinese models. Behind the scenes, Axios says this fight has been simmering for a while. US Commerce has been considering blacklisting Chinese AI labs since last year, and the White House had a draft executive order that would've made US companies liable for running Chinese models. All of it stalled, but Kimi's release seems to have brought it back to life. Jensen Huang split the difference on the "they stole our work" question: "Distillation, learning from AI, learning from other sources of knowledge, is fundamental to intelligence." Read the incentives Granted, all this commentary is not coming from a neutral party. Nvidia's China revenue is basically zero right now, down from a business Huang once said could be worth $50 billion a year. Huang has spent two years arguing against export controls, so of course he's going to say demand for AI is elastic and that restrictions can backfire. That said, he has also endorsed keeping Blackwell and Rubin out of China's hands. Inside Wistron's AI infrastructure manufacturing facility in Fort Worth, Texas. The good news is that his core claim could soon be verified. Every number on Kimi K3 so far comes from Moonshot itself or from early API testing. The full weights go public on July 27, and at that point the benchmarks either hold up under independent testing or they don't. Anthropic has accused Moonshot of training on 3.4 million Claude conversations earlier this year, though analyst Nathan Lambert's take is that even if some of that happened, it's not enough to explain how good K3 actually is. Separately, Epoch AI estimates the gap between the best open and closed models is now down to around three months. If that gap keeps shrinking, the real question stops being whether US companies should be allowed to use Chinese models, and starts being whether banning them would even do anything once the weights are already sitting on servers everywhere.

China's Moonshot AI Unveils Kimi Model, Threatening America's Lead

Meaghan Tobin reported from New York and Cade Metz from San Francisco. The contest between China and the United States for supremacy in artificial intelligence escalated on Friday when the Chinese start-up Moonshot AI released a new A.I. model that appeared to narrow the lead held by well-funded American competitors. Moonshot said that the model, Kimi K3, was the world's largest open-source A.I. system, allowing anyone to use, modify and build on it freely. The company said that Kimi K3 performed as well as leading models from OpenAI and Anthropic at some key tasks. The release coincided with an address by Xi Jinping, China's leader, in which he outlined an ambitious vision for global A.I. development that cast China as the champion of an open approach to the technology. "A.I. development should not be a solo performance by a single country but a symphony of international cooperation," Mr. Xi said. China was "ready to work with all parties to seize the opportunities of A.I. development," he said. The market reaction was immediate. Stocks fell on Friday, with the Nasdaq dropping about 1 percent as investors sold shares of companies that make the computer chips needed to power A.I., including Nvidia and Intel. The developments underscored the geopolitical stakes of the U.S.-China contest for leadership in what many see as the defining technological race of our time. President Trump has repeatedly framed maintaining America's lead in artificial intelligence as a strategic imperative for preserving the country's economic and national security. The emergence of A.I. as a foundational technology with far-reaching economic and military implications has pushed the United States to slow China's progress by restricting its access to the advanced chips needed to train cutting-edge A.I. systems. The release of a free, open-source Chinese A.I. model that could rival the performance of costly, computing-intensive systems from Silicon Valley's best-funded companies has reignited concerns over whether the industry's enormous spending spree on data centers is justified. Over the past 18 months, Chinese technology companies have released a steady stream of open-source A.I. models that approach the performance of their American rivals, while claiming to require fewer computing resources. That combination of strong performance and lower costs has made Chinese A.I. systems the preferred choice for many developers worldwide. Moonshot's latest release adds to that momentum and raises the prospect that a Chinese company could soon match -- or even surpass -- the cutting-edge models built by leading American A.I. firms, which have raised hundreds of billions of dollars to build models that are closed and expensive to operate. According to benchmarks run by Vals AI, an independent company that evaluates the performance of A.I. models, Kimi K3 performs just below Anthropic's Fable 5 model while outperforming OpenAI's flagship GPT-5.6 Sol model, said Rayan Krishnan, Vals AI's chief executive. The latest A.I. models from Anthropic have become so powerful that their capabilities -- including identifying vulnerabilities in critical infrastructure such as banking networks and power grids -- have alarmed governments around the world. Graham Webster, a Stanford University professor who studies China's open-source A.I. ecosystem, said it did not matter much whether a model was at the absolute cutting edge, because the performance gap between leading American and Chinese models had narrowed. "Everyone has a different number about how many months China is behind the U.S., and the estimates cluster around six months," he said. "That is not much of a lead." Moonshot said it would release the full details of Kimi K3 at the end of the month. Samm Sacks, a senior fellow at the Johns Hopkins School of Advanced International Studies who attended Friday's A.I. conference in Shanghai where Mr. Xi spoke, said that many experts believe it's only a matter of time before a Chinese company releases a freely available model that matches Anthropic's most advanced systems. Such a breakthrough, Ms. Sacks said, would raise difficult questions for regulators trying to contain powerful A.I. technology once it has been released. Last month, another Chinese start-up, Z.ai, released a model called GLM-5.2 that was nearly as powerful as Anthropic's Fable 5 model, the leading American system. And unlike Fable, this Chinese model was widely available. Many software developers and start-ups in Silicon Valley quickly adopted the model, largely because it was much cheaper than the leading American systems. It arrived just as U.S. businesses realized they had to find ways to cut down their A.I. spending. Kimi K3 is even more powerful than GLM-5.2, according to benchmarks run by Vals AI. The model arrives as the Trump administration is considering regulating the technology. Many Silicon Valley executives have begun to support such regulation, saying that leading technologies can enable malicious cyberattacks and potentially help build biological weapons. Xinyun Wu contributed research from Taipei.

China's Moonshot AI claims Kimi K3 can rival OpenAI and Anthropic

Chinese AI start-up Moonshot has unveiled a massive new artificial intelligence model it says can rival top American firms. The company launched Kimi K3, containing 2.8 trillion parameters, which serves as a measure of an AI's scale and processing power. Kimi K3's full capabilities - coding, knowledge work, and reasoning - will be known when it is released as an open-source model on 27 July. The sudden breakthrough suggests that China's tech prowess is rapidly narrowing the capabilities gap, upending long-held assumptions in the West that Chinese developers trail their American peers. Its arrival later this month will make it the world's first open-source model in the three-trillion-parameter class that can be freely downloaded, run and customised by outside developers. The release comes at a highly sensitive moment for the global technology sector, just weeks after the US government abruptly forced American developer Anthropic to temporarily withdraw its flagship Fable and Mythos models due to severe cybersecurity concerns. While Washington has since lifted those restrictions, the initial move highlights how the US government now views advanced AI software as critical national infrastructure, labelling frontier models as vital national security assets subject to strict export controls. However, the rapid arrival of Kimi K3 suggests Chinese firms are successfully bypassing these regulatory barriers and advancing independently despite US restrictions on hardware sales. Heavily backed by domestic tech giants Alibaba and Tencent, Moonshot has quickly risen to the forefront of China's generative AI ecosystem. In a statement the company said that K3 stands as Moonshot AI's "most capable flagship model to date". Unlike closed, proprietary American systems from OpenAI or Anthropic, Kimi K3's open nature allows global users to modify the system for advanced reasoning and complex software development. Moonshot AI noted that the system is uniquely built to operate with "minimal human supervision" to sustain tasks such as engineering and coding. Third-party evaluations from Artificial Analysis and Arena.ai show the model performing on a par with leading models in the US, such as OpenAI's GPT and Anthropic's Claude. In these independent benchmarks, Kimi K3 ranked first in web interface engineering, outperforming Anthropic's Fable system in blind human-preference tests. While the system's massive size means running it locally requires significant computing equipment, making it open-source could heavily disrupt Silicon Valley's commercial models. The announcement had an immediate impact on shares in Moonshot's domestic competitors Zhipu and MiniMax, which tumbled sharply in Hong Kong by about 27% and 16% respectively.

Jensen Huang calls the AI jobs panic 'complete nonsense'

The most powerful man in AI thinks the doomers, some of them his own customers, have it backwards. In a long interview with Axios, Jensen Huang called fears of mass job losses "complete nonsense." His argument rests on a neat distinction: your task is not your job. Jensen Huang has heard the warnings. He is not buying them. The Nvidia chief runs the world's most valuable company on the back of the AI boom. So when he calls the darkest predictions about that boom "complete nonsense," it is worth asking whether he is right, or just talking his book. In a wide-ranging Axios interview with Mike Allen, Huang took on two claims at once. That AI will end humanity, and that it will wipe out half of American jobs. Both are "complete nonsense," he said, and "all of the evidence points exactly to the opposite." His real argument is subtler than the soundbite. It is not that no jobs change. It is that people keep confusing a task with a job. Don't mistake your task for your job The line is one Huang says he gives to young people. "Don't mistake your task for the job." A radiologist's task is reading scans, and AI can do much of that. But the job, he argues, is ending suffering and guiding patients. That does not vanish when the scan-reading is automated. Some roles will still go. Call-centre work, he granted, is "very likely to be highly automated." The judgement around it, he thinks, stays human. He came with numbers. The number of radiologists is rising, by about 20%, he said, because AI lets each one see more patients, a claim others have backed. Paralegals are up around 10%. Manufacturing jobs have grown some 50% in recent years, he claimed, as the industry races to build data centres. His reframe of the fear is blunt. "AI is not going to destroy all of our jobs," he said. "Someone who uses AI is going to take your jobs." A jab at his own peers The pointed part was his target. Asked why Asia is so much sunnier about AI, Huang said the doomers "spend too much time theorizing about these science fiction outcomes," which "makes them sound smart." Pressed on whether he meant fellow AI bosses, he softened. The aim was clear enough. Talk of a singularity, of a simulation, of machine consciousness, is "all made up." "It's fine to warn people," he added. "It's absolutely inappropriate to make things up." He took the same message to Washington. AI is not a "100-metre dash" with a single winner, he said, and policymakers risk over-correcting if they "fall for these narratives." His advice to the Trump administration was to talk to many scientists, not one or two. He even waved off the idea of the government taking a stake in Nvidia as "unnecessary." The optimist's ledger Huang's brighter vision is expansive. He expects a trillion AI agents running at once. He says the "ChatGPT moment" for robots has already arrived, with useful machines perhaps three to four years away. An AI bubble "will come someday," he allowed, "just not today," and probably not within five years. It is a bracing counter-story, and Huang has every reason to tell it, because more AI means more Nvidia chips. The job figures are his own framing, and rivals such as Anthropic's Dario Amodei still forecast the opposite. Yet his central distinction is hard to wave away. Automation has reshaped work before without ending it, and "someone who uses AI is going to take your jobs" is a more honest worry than a robot uprising. The doom, Huang bets, is a story. The real question is whether people believe him before they act on the other one.

Choose Your Fighter: Nvidia CEO and Jim Cramer Offer Dueling Visions of AI's Future

The release of a powerful open-source AI model from China has once again split the tech industry into two camps: the evangelists who believe good tech should be accessible and the protectionist hawks who believe anything foreign is a threat. It must simply kill Jim Cramer, host of CNBC's Mad Money and a guy who has never once been deterred by being both loud and wrong, to find himself on the opposite side of Nvidia CEO Jensen Huang, given how much he loves to tell people to smash the buy button on the chipmaker. But Cramer is a China hardliner, refusing to believe anything coming from the country could be anything but compromised. Huang, meanwhile, thinks access to good, cheap AI models is a net positive regardless of where they come from. In an interview with Axios, Huang said plainly that the Chinese models like the recently released Kimi K3 are "excellent" and that "open-source models that are excellent should be used." That's a pretty simple equation to follow: If model equals good, use the good model. He also dismissed the idea that good models from China are somehow a threat to American labs like OpenAI and Anthropic. "There's no scenario where China runs U.S. companies off the road," he told the publication. "Zero possibility." You'd be hard-pressed to get Cramer to believe that. Ol' Jim drew a line in the sand on X, stating explicitly, "We must NOT let our companies use these Chinese models to save a few bucks. OpenAI and Anthropic are correct. This is vital national security," he said. "I respect the Chinese people greatly but these companies are run by the PLA [People's Liberation Army] for heaven's sakes." That last claim seems to be baseless. Moonshot AI, the maker of the Kimi AI model, does have state-linked investors, but there's no real evidence of direct military ties. Rival lab DeepSeek has been accused of lending its capabilities to the Chinese military, and members of the US Congress have pushed for it to be designated as a Chinese military company -- but you'd be pressed to figure out what is different about what DeepSeek is doing for the Chinese government than what the Trump administration has pushed Anthropic and other American AI labs to do for the Pentagon. Huang doesn't seem impressed by people like Cramer ringing the alarm bells. He told Axios that it's a "misconception" that Chinese models provide a "backdoor" to the nation's government, and said that companies that want to use an open-source model like Kimi can control access when they deploy it. He also stated the obvious: that the open-source nature of the models makes them much easier to check for security vulnerabilities. Nvidia's top exec also called out the fact that Cramer's pals in the finance world keep lighting their hair on fire for no reason. "The market misunderstood the impact of DeepSeek the first time," he told Axios, and they've "misunderstood the impact of Kimi again this time." Huang's case is basically that cheaper models are good for business all around, increasing demand for data centers and computing power while also being less burdensome on all those chips that get deployed. "Free AI should be great for chips. Free AI should be great for data centers," he said. Of course, there's a strange thing that happens with the "free enterprise" folks like Cramer when an outside player threatens to eat into the American share of the pie. Suddenly the market actually does need regulating, and companies perceived as "threats" need to be locked out. If Cramer has his way, Kimi will be the BYD of AI -- cheap, good, and inaccessible to Americans.

World's first 3-trillion model from China does weeks of work in hours

Moonshot AI has unveiled Kimi K3, a 2.8 trillion-parameter open-source AI model described as the world's first open 3T-class model. Hailed as the largest open-source AI model released to date, Kimi K3 follows Zhipu AI's release of GLM-5.2 one month earlier. Efforts are intensifying among Chinese AI developers to narrow the gap with leading U.S. AI companies, according to the South China Morning Post. Designed for complex, long-horizon workflows, Kimi K3 features native vision capabilities and a one-million-token context window. This capability allows for longer coding sessions, large repositories, and workflows combining text, images, and other information, according to Moonshot. The company positioned Kimi K3 to support scientific research workflows. It can create research reports with interactive visualizations, scientific analyses, and editable presentations. Its Widgets and Dashboard features allow users to create persistent, interactive workspaces. The model completed a task in about two hours that would typically require one to two weeks of work by an experienced researcher. It is no wonder that Kimi K3 demonstrated "frontier-level performance across our evaluation suite, consistently outperforming other tested models," as the company reported. However, it "still trails the most powerful proprietary models, Claude Fable 5 and GPT 5.6 Sol," as per the blog post. Want to try it? You can. In its launch blog, Moonshot AI said Kimi K3 was built using its Kimi Delta Attention and Attention Residuals architecture, along with a Mixture-of-Experts framework that activates 16 out of 896 experts. The company stated that these updates improve scaling efficiency by approximately 2.5 times compared to Kimi K2. Kimi K3 achieved "open frontier intelligence" with 2.8 trillion parameters and was designed for "long-horizon coding, knowledge work and reasoning tasks," according to the South China Morning Post. Moonshot's benchmark results showed Kimi K3 outperforming several tested systems, including OpenAI's GPT-5.5, Anthropic's Claude Opus 4.8 and Zhipu AI's GLM-5.2. Moonshot demonstrated the model's capabilities on its blog. The company showed Kimi K3 optimizing GPU kernels, developing a GPU compiler called MiniTriton, designing a prototype AI chip using open-source electronic design automation tools, and completing a computational astrophysics workflow. The model reproduced the I-Love-Q universal relations in computational astrophysics by reviewing and cross-validating more than 20 papers, implementing the numerical pipeline, evaluating more than 300 equations of state, generating more than 3,000 lines of Python code, and producing an interactive HTML dashboard. With a market size set to grow from $539.5 billion to $3,497.3 billion by 2033, AI's global takeover appears undeniable, if not unstoppable. More than 50% of U.S. jobs are projected to be significantly altered by artificial intelligence within the next two to three years, according to the Boston Consulting Group. Kimi K3 has officially entered the market, as China aims to compete with U.S. companies. Kimi K3 is available now through Kimi.com, Kimi Work, Kimi Code, and the Kimi API, according to Moonshot AI's blog. The AI model launches with maximum reasoning effort enabled by default, while low- and high-effort modes will be introduced in future updates. The full model weights will be released by July 27, 2026, alongside a technical report covering the model's architecture, training process, and evaluations. Users can try Kimi K3 now through Moonshot's platforms.

China's 2.8 trillion-parameter Kimi K3 beats Claude Fable 5 in Frontend Code Arena -- the largest open-weight AI model ever, as China works around U.S. compute limits

Moonshot's own disclosures point to export-grade Nvidia silicon and an unnamed alternative GPU vendor. Beijing-based Moonshot AI has released Kimi K3, a 2.8 trillion parameter model that the company describes in its technical blog as the world's first open 3T-class system and the largest open-weight AI model to date. Moonshot said K3 still sits behind Anthropic's Claude Fable 5 and OpenAI's GPT 5.6 Sol on overall performance, but it outperformed every other model in the company's evaluation suite, including Claude Opus 4.8 and GPT 5.5, across coding and agentic benchmarks. The model has a 1 million token context window, native vision, and activates just 16 of its 896 experts per token, roughly 1.8% of the pool. Full weights are due by July 27. Arena ranked K3 first in its Frontend Code evaluation at 1,679 points, ahead of Fable 5, in blind developer testing. API pricing is $0.30 per million cache-hit input tokens, $3 per million on cache misses, and $15 per million output tokens. Kimi K2 launched a year ago at $0.60 per million input tokens, so uncached K3 input costs five times as much. Moonshot claims roughly a 2.5x improvement in scaling efficiency over Kimi K2, attributed to two architectural changes: Kimi Delta Attention, a hybrid linear attention scheme, and Attention Residuals, which change how information moves between layers. Quantization-aware training starts at the supervised fine-tuning stage, using MXFP4 weights and MXFP8 activations, a combination Moonshot says it chose for broad hardware compatibility. Bank of America analysts led by Alex Liu said in a note cited by CNBC that K3 shows large-scale pre-training plus architectural work can still deliver step-change gains for flagship Chinese models despite compute constraints. Moonshot's kernel optimization benchmark ran on Nvidia's H200, and what the blog identifies only as a "GPGPU from an alternative vendor," which the company didn't name. MiniTriton, a Triton-like compiler K3 built from scratch, is charted against Triton on an Nvidia L20, the cut-down Ada-based card sold into China under U.S. export rules. Moonshot recommends serving K3 on supernodes of 64 or more accelerators, keeping expert-parallel traffic inside one high-bandwidth domain. The blog doesn't say where the H200 hardware is; Congress passed a bill in January to close the offshore cloud rental loophole that gave Chinese firms remote access to restricted accelerators. In one case study, K3 spent a single 48-hour autonomous run designing a simulated inference chip for a nano model built on its own architecture, using open-source EDA tools and the Nangate 45nm library. The design closed timing at 100 MHz within 4mm squared, packed 1.46 million standard cells and an INT4 MAC array, and sustained more than 8,700 tokens per second of simulated decode. At the moment, every published K3 number is a claim made by Moonshot-reported or drawn from API access and can't be verified until the weights are made public on July 27. Anthropic accused Moonshot in February of using 3.4 million Claude exchanges to train its models through distillation, and K3 now benchmarks within a few points of the models named in that complaint. Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.

US threatens to sanction Chinese AI; Huang pushes back

Washington wants to sanction China's open-source AI models, calling them theft. Nvidia's Jensen Huang, the man who sells the chips underneath all of it, says that is exactly backwards. "There's no scenario where China runs US companies off the road," he told Axios. "Zero possibility." On Tuesday, Treasury Secretary Scott Bessent went on television with a warning for China. On the same day, the chief executive of the company that powers the entire AI boom went the other way. The threat Bessent said the US will scrutinise Chinese open-source models for stolen intellectual property, as Bloomberg first reported. It could then hit their makers with sanctions. "This administration supports open source models, but what we do not support is IP theft," he said. His evidence is watermarks. "We are finding watermarks of our US large language models on many Chinese models," he said. The technical term is distillation, training one model on another's outputs. Bessent called it theft, and promised action within weeks. What set it off The trigger is Kimi K3, an open-weight model from China's Moonshot AI. It matches or beats OpenAI and Anthropic's best on some benchmarks, and costs far less. Its arrival sparked a chip-stock selloff and a fresh bout of Washington panic. US labs have accused Chinese rivals of siphoning their models for months. Huang's rebuttal Nvidia's Jensen Huang, speaking to Axios at a new chip plant in Texas, called that panic misplaced. "These Chinese models are excellent," he said. "Open-source models that are excellent should be used." American firms, he argued, should "absolutely" be allowed to run them. His logic is commercial. Cheaper, free models put AI in more hands. More AI means more demand for the chips, data centres, and power Nvidia sells. "Free AI should be great for chips," he said. He also rejected the fear that a downloaded Chinese model is a "backdoor" to Beijing. Firms can wall them off in secure sandboxes, he said. The awkward part Not everyone accepts the theft charge. Distillation is common, and US labs do it too. "We know distillation to be a very small factor," Hugging Face chief Clem Delangue told TechCrunch. It is, he added, "a practice that everyone is doing, including companies in the US." There is an irony too. Anthropic, one of the loudest accusers, just had its own $1.5bn settlement approved over pirated books used to train Claude. Huang went further, urging the government to stop restricting Anthropic's own Mythos model. "Let Anthropic run," he said. What happens next Bessent will lead the US side at AI talks with China in September. The distillation fight will be on the table. For now the split is stark: the government wants to fence Chinese models out, while the world's most valuable chipmaker wants them let in.

China Just Dropped Another Bomb on America's Frontier AI Companies

Alibaba-backed Chinese artificial intelligence startup Moonshot just unveiled its latest model, Kimi K3, and it's already sending shockwaves through the industry, with some benchmarks showing the model outperforming Anthropic and OpenAI's best offerings. The model packs 2.8 trillion parameters, which Moonshot says would make it the largest open-weight model released to date once its weights become available by July 27. In a blog post, the company acknowledged that K3's overall performance still trails Claude Fable 5 and GPT-5.6 Sol. Its internal evaluations nevertheless place it close to both models on several tasks, while independent testing by Artificial Analysis ranks it immediately behind the leading proprietary systems on its Intelligence Index and real-world work evaluations. On Arena.ai's front-end development leaderboard, K3 even ranks above the two most powerful models, marking a 17-place jump from the company's previous model, Kimi K2.6. Arena's CEO, Anastasios Angelopoulos, said Kimi K3 "may be the single biggest release of the year" and "the moment that OSS Chinese models have surpassed US models," in a post on X. It's a remarkable achievement, especially for an open-source model. The results challenge the assumption that China's leading AI labs remain several months behind their American competitors. Anthropic just released Fable 5 last month, while OpenAI's GPT-5.6 (and its three tiers, Sol, Terra, and Luna) just dropped last week. "Kimi k3 is a big moment with multiple implications for the entire industry," Trump's former senior White House policy advisor on AI, Sriram Krishnan, said in a post on X. The last time something like this happened, aka when a Chinese AI lab released a cheaper model that proved competitive with American alternatives, was when DeepSeek released R1 back in January 2025. Following that release and its reception, the market reaction helped wipe roughly $1 trillion from global technology stocks. Meanwhile, the model's success raised major national security concerns across Washington D.C., and partially informed the Trump administration's hard-line stance on advanced tech exports to China. Moonshot's release also comes only a few months after Anthropic accused the company, along with other Chinese AI companies DeepSeek and MiniMax, of violating their rules to "illicitly" extract the capabilities of its model Claude and use that to improve their own models. The process is called "distillation," and it's fairly common in the industry, but the Trump administration has deemed it "adversarial" and vowed to crack down on it. K3 arrives amid heightened scrutiny of the U.S.-China AI race and growing national-security concerns around frontier models. Its release is likely to renew debate in Washington over export controls, distillation, and whether restrictions on Chinese labs are slowing their progress at all.

Axios interview: Jensen Huang is AI's anti-doom evangelist

Why it matters: Huang is influential with the Trump administration, putting him at the center of a debate with enormous consequences for America's AI future. Huang railed against AI doomerism in an interview for our "Behind the Curtain" video series -- implicitly challenging fellow tech CEOs who have emphasized the technology's potentially catastrophic risks. * "The fact that this is going to be the end of humanity -- it's complete nonsense," Huang told us. "The fact that this is going to destroy half of the American jobs is complete nonsense." * Scaring workers and companies away from using AI, he said, poses a greater danger. * We sat down with Huang in Fort Worth, Texas, at the grand opening of a new phase of an advanced manufacturing plant for Taiwan-based Wistron, which makes AI infrastructure systems for Nvidia. 🔬 Zoom in: Huang called on policymakers to consult more than "one or two" CEOs and avoid restricting the technology based on scenarios that have not materialized. * He also suggested some companies invoke safety concerns to secure favorable regulations: "Some of the companies hope that the government would be helpful in creating regulations to their advantage." * Asked whether he feared the administration could overcorrect, Huang said yes. Between the lines: Anthropic and OpenAI have pushed Washington to take the most advanced systems and their potential risks seriously. Critics say those campaigns could also entrench their market position. * Huang didn't name either company, and Nvidia is a major partner and supplier to both. * Anthropic CEO Dario Amodei, in particular, has been at the center of high-stakes fights with the Pentagon and White House over how the government should use and deploy advanced AI. * Washington is also deciding what to do about powerful open-source models emerging from China. 🖼️ The big picture: Officials are accusing Chinese firms of stealing from American models to build their own as they rapidly become more competitive with OpenAI and Anthropic, Axios' Maria Curi writes. * Treasury Secretary Scott Bessent on Wednesday threatened sanctions against Chinese firms conducting what he called "industrial-scale distillation attacks." * Earlier that day, the White House's top science official accused Chinese startup Moonshot AI of distilling Anthropic's technology to build its fast-rising Kimi model and said it likely used Nvidia chips to do it.

Moonshot reveals new AI model, and it's a big surprise -- here's why Kimi K3 is a threat to the likes of OpenAI

* Moonshot has launched a new AI model, Kimi K3 * It's surprisingly powerful, with the Chinese AI firm claiming it outguns most US rivals, save for a couple of exceptions * Kimi K3 is open weight by nature, which poses a further threat to the likes of OpenAI and Anthropic Moonshot, one of the emerging Chinese AI giants, has just revealed a new AI model which is seemingly up there with the likes of ChatGPT and Claude. Bloomberg reports that Moonshot's new Kimi K3 model can equal the best that the US has to offer, at least based on the company's own benchmarking. Seemingly it outguns all rival AIs save for Claude Fable 5 (from Anthropic) and GPT-5.6 (from OpenAI). Kimi K3 is a model with 2.8 trillion parameters, Bloomberg tells us, and Artificial Analysis ranked it ahead of Anthropic's Opus 4.8 on some benchmarks. Moonshot also claims it beats Chinese rival Z.AI for coding tasks, and overall, the performance of the new model has caught the market by surprise. Moonshot notes in a blog post that Kimi K3 is the "world's first open 3T-class model, designed for frontier intelligence across long-horizon coding, knowledge work, and reasoning." Kimi K3 is available to use now. Bloomberg quotes Leonid Mironov, a portfolio manager at Gavekal Capital, as saying: "In my use, it's clearly the best Chinese model ever," noting that it's "brilliant" no less. Analysis: a weighty threat This is a threat to the big US players in the AI market for several reasons. The key difference with Kimi K3 is that it's what's known as an "open weight" model, meaning anyone can grab the model to run it for themselves - from July 27, when the weights are released - without paying anything. It's not the same as open source, though, as while you can get the model, what you don't get to do is peek behind the scenes at how the model was trained (and on what data). The other caveat is that running Kimi K3 takes some extremely powerful hardware; but nonetheless, for firms with the substantial wherewithal to do that, the pre-trained model is there for the taking at no cost. So, you can imagine how this open weight approach is threatening to the AI behemoths in the US (and this is presumably the point of going this way for the Chinese rival). What will also be a concern to the likes of OpenAI and Anthropic is that Moonshot is attacking one of the most lucrative aspects of AI, with Kimi K3 being pushed for its coding skills. However, Moonshot is charging a lot more than Chinese rivals for those who want to use it, and in fact, it's priced around Claude Sonnet levels, so on a par with the current cutting-edge (frontier) AI models. That in itself is a signal of the quality on offer here, and why Kimi K3 has raised quite a few eyebrows. As the competition around AI heats up, there are also concerns about whether that means safeguards will be increasingly overlooked in favor of faster development and progress (which has been a consistent source of worry for many as it is). Adding to all the controversy are accusations of AI theft leveled by the US State Department at Chinese firms earlier this year, Moonshot included. Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.

China's Moonshot AI releases Kimi K3, the largest open-source model ever, rivaling top U.S. systems

Moonshot AI, the Beijing-based artificial intelligence startup backed by Alibaba, on Thursday released Kimi K3 -- a 2.8-trillion-parameter model that the company says is now the largest open-source AI model in the world, and one that benchmarks show performs neck-and-neck with the most powerful proprietary systems from Anthropic and OpenAI. The release, timed to land just ahead of the 2026 World Artificial Intelligence Conference in Shanghai, is a dramatic escalation in the global AI arms race and a watershed moment for the open-source AI movement. It also marks a remarkable comeback for a company whose market position had eroded significantly over the past 18 months following DeepSeek's meteoric rise. Full model weights are scheduled to be released on July 27, according to details shared by researchers who reviewed the company's technical documentation. If you want to take Kimi K3 for a spin right now, you can -- just head to kimi.com, sign up with a Google account or phone number (no credit card required), and start chatting with what may be the most powerful open-source model ever built. Inside the architecture that powers the world's largest open-source AI model Kimi K3 is a frontier-class large language model with 2.8 trillion total parameters -- roughly 75 percent larger than DeepSeek's V4 Pro, which the company's own timeline chart shows at approximately 1.6 trillion parameters. The model features a 1-million-token context window, native visual understanding capabilities, and an always-on reasoning mode that the company calls "thinking mode." The model is built on two key architectural innovations developed internally at Moonshot AI: Kimi Delta Attention, a hybrid linear attention mechanism, and Attention Residuals, which the company describes as a drop-in replacement for residual connections that delivers consistent scaling gains. Both techniques were previously published as open research by the Moonshot team on GitHub. On the API side, Kimi K3 is compatible with the OpenAI SDK, lowering the integration barrier for developers already building on OpenAI or Anthropic toolchains. The model is priced at $3 per million input tokens and $15 per million output tokens, with cached input tokens dropping to just $0.30 per million -- pricing that positions it roughly in line with mid-tier offerings from Western labs, but at a performance level the company claims approaches the top of the market. A promotional top-up rebate running through August 12 offers up to 30 percent back in vouchers for API credits of $1,000 or more. As Xinhua reported, a Moonshot AI executive explained the significance of the parameter count in simple terms: parameters are like neural connections in the human brain, and nearly 3 trillion of them means the model can "store more knowledge and patterns in its brain, understand more, think deeper, and answer more accurately." Benchmark results show Kimi K3 trading blows with Claude and GPT at the top of the leaderboard The benchmark results, drawn from public leaderboard data and a private evaluation by analytics firm Artificial Analysis, tell a striking story. On GDPval-AA v2, a benchmark measuring real-world tasks across 44 occupations and 9 major industries, Kimi K3 scored 1,687 -- placing it third overall, behind only Claude Fable 5 Max (1,815) and GPT-5.6 Sol Max (1,747.8), and ahead of Claude Opus 4.8 (1,600). On AA-Briefcase, a private agentic benchmark from Artificial Analysis designed to test long-horizon knowledge work, K3 climbed to second place with a score of 1,527 -- beating GPT-5.6 Sol Max (1,495) and trailing only Fable 5 Max (1,587). Perhaps most impressively, K3 achieved a state-of-the-art score of 91.2 out of 100 on BrowseComp, a benchmark for long-horizon, high-difficulty information seeking. The company says it accomplished this in a single-agent setup using its 1-million-token context window, without any context compression or additional context management techniques -- a feat that suggests raw context length, when paired with strong retrieval capabilities, may be more powerful than elaborate multi-agent workarounds. As one widely followed AI commentator put it on social media: "Open source is no longer lagging six months behind Western closed-source models. Read that again, and think about what it all means." That observation captures the significance of the moment. For much of the past three years, open-source models have typically trailed their proprietary counterparts by a meaningful margin. Kimi K3 appears to have closed that gap almost entirely. How a 48-hour autonomous chip design demo reveals Moonshot's real ambitions Beyond raw benchmarks, Moonshot AI showcased a proof-of-concept that may be even more revealing of K3's capabilities and the company's strategic direction. In a demonstration documented in the company's technical materials, Kimi K3 was tasked with designing a physical chip to run a nano-scale version of itself. Over 48 hours of continuous autonomous agent operation, K3 independently completed the chip's full construction pipeline -- from architectural design through optimization and verification -- using open-source electronic design automation tools. The result was a tiny but functional chip design, just 4 square millimeters, that achieved timing convergence at 100 MHz and could decode more than 8,700 tokens per second in simulation. This is not a production chip. It is a demonstration of what Moonshot AI clearly views as the next competitive frontier: long-range autonomous agent capabilities. The ability to sustain coherent, multi-step technical work over a 48-hour window -- reading documentation, making design decisions, running verification loops, and iterating on failures -- represents a qualitative leap beyond the kind of single-turn question-answering that defined the first generation of large language models. The company also highlighted a case in computational astrophysics, where K3 reportedly reproduced the universal I-Love-Q relation -- a complex calculation that typically takes a senior researcher one to two weeks -- in approximately two hours, reading and cross-validating more than 20 papers and implementing a complete numerical pipeline along the way. Moonshot AI's fall and rise tells the story of China's brutal AI market To understand why Kimi K3 matters, you need to understand where Moonshot AI was 18 months ago -- and how far it fell. Founded in 2023 by Yang Zhilin, a Tsinghua University graduate who previously conducted research at Google and Meta, Moonshot AI quickly became one of China's most prominent AI startups. The company gained early traction in 2024 when users flocked to its Kimi platform for its long-text analysis capabilities and AI search functions. By early 2026, it had raised roughly $1.5 billion across multiple rounds, with its valuation climbing from $2.5 billion to $4.3 billion and the company reportedly seeking a new round at $5 billion. Then DeepSeek happened. The release of DeepSeek's low-cost R1 model in January 2025 disrupted the entire Chinese AI landscape, and Moonshot AI was among the hardest hit. Kimi, which had ranked third in monthly active users in China, slid to seventh. The company's strategic pivot to open-source models -- beginning with Kimi K2 in July 2025 and accelerating with K2.5 in January 2026 -- was in large part an effort to reclaim relevance. Kimi K3 is the culmination of that effort -- and the sheer scale of the model suggests that Moonshot AI has been planning this move for some time. Training a 2.8-trillion-parameter model requires enormous computational resources and months of preparation, which means the architectural and infrastructure decisions behind K3 were likely locked in well before the model reached the public. Why open-sourcing the world's biggest model is a geopolitical chess move The decision to release K3's full weights on July 27 is strategically significant and worth parsing carefully. The company's own timeline chart of open-source frontier model scale positions K3 as a dramatic outlier, towering above competitors like DeepSeek (1.6T), Xiaomi (1.02T), and Alibaba (397B). By releasing the world's largest open-source model, Moonshot AI is making a bid to become the center of gravity for the global open-source AI developer community. This follows a broader trend among Chinese AI companies. As Reuters noted, open-sourcing allows companies to "showcase their technological capabilities and expand developer communities as well as their global influence, a strategy likely to help China counter U.S. efforts to limit Beijing's tech progress." DeepSeek, Alibaba, Tencent, and Baidu have all released open-source models. But none have released anything at this parameter count. For enterprise technology leaders, the implications are concrete. A 2.8-trillion-parameter open-source model that performs at near-frontier levels creates new options for companies that want to fine-tune, self-host, or build proprietary systems on top of a capable base model -- without being locked into API contracts with OpenAI or Anthropic. The trade-off, of course, is that running a model of this size requires substantial GPU infrastructure. Inference at 2.8 trillion parameters is not something that runs on a single server rack. That said, Moonshot AI has signaled awareness of this challenge. Its Mooncake project, which won the Best Paper award at FAST 2025, pioneered KV-cache-centric disaggregated serving for large language models -- an architecture designed specifically to make inference at extreme scale more practical and cost-efficient. Kimi Code and a three-tier model lineup form the foundation of Moonshot's enterprise play Alongside K3, Moonshot AI continues to invest heavily in its coding agent ecosystem. Kimi Code, the company's open-source coding tool that competes with Anthropic's Claude Code and Google's Gemini CLI, received two major updates on the same day as K3's launch -- versions 0.25.0 and 0.26.0 -- adding features like expanded subagent tooling, background task management, and security fixes. The Kimi Code CLI has accumulated over 3,100 stars on GitHub and features integration with VSCode, Cursor, and Zed. The latest release expanded the "coder subagent" tool set to include background tasks, todo lists, plan mode, skill invocation, and nested agents -- effectively turning the coding agent into a multi-layered autonomous system capable of managing complex software engineering projects with minimal human intervention. This is not incidental. Coding tools have become a critical revenue driver for AI labs. As Anthropic disclosed in January, Claude Code reached $1 billion in annualized recurring revenue. By building Kimi Code as an open-source alternative that defaults to Kimi's own models -- but supports other providers -- Moonshot AI is positioning itself to capture developer workflows and, eventually, enterprise contracts. The company's model lineup now includes three tiers: K3 as the flagship ($3/$15 per million tokens for input/output), K2.7 Code as a specialized coding model ($0.95/$4), and K2.6 as a general-purpose option ($0.95/$4). All three support context windows of 256,000 tokens or above, with K3 offering the full 1-million-token window. Context caching is automatic -- no cache ID, TTL, or extra parameter is required -- a small but meaningful developer-experience advantage over competitors that require explicit cache management. What Kimi K3 means for the future of enterprise AI and the global model landscape Kimi K3's release forces a recalibration of several assumptions that have guided enterprise AI strategy. The performance gap between open-source and proprietary models has functionally closed at the frontier. If K3's benchmark numbers hold up under independent evaluation -- and particularly once the open weights are available for community testing on July 27 -- it will be difficult for closed-source providers to justify premium pricing purely on the basis of capability. The locus of AI innovation, meanwhile, continues to shift. China's AI ecosystem, which many Western observers questioned after early struggles with chip export restrictions, has now produced a model that competes with the best systems from companies with direct access to Nvidia's most advanced hardware. The architectural innovations behind K3 -- particularly the hybrid linear attention mechanism -- suggest that algorithmic efficiency may matter as much as raw compute. And the agentic capabilities demonstrated by K3 -- chip design, multi-week research compression, long-horizon information seeking -- point toward a future where AI models are not just answering questions but autonomously executing complex, multi-day projects. For enterprises evaluating AI investments, this shifts the value proposition from "productivity copilot" to "autonomous technical workforce." Xinhua, China's state news agency, framed the release as a national milestone, reporting that K3 "marks a new step forward in the development of China's artificial intelligence models." Liu Tieyan, dean of the Zhongguancun Academy in Beijing, was quoted as saying that a wave of Chinese open-source models has moved from isolated breakthroughs to collective advancement, providing "new solutions and new paths" for global AI development. Just two years ago, Moonshot AI was a scrappy startup named for the audacious problems it hoped to solve. Eighteen months ago, it was a cautionary tale about how quickly a market darling can lose its footing. Today, it is the maker of the world's largest open-source AI model -- one that can, given 48 hours and an internet connection, design a chip to run itself. The frontier, it turns out, is not a place. It is a race. And the field just got a lot more crowded.

Jensen Huang says U.S. firms should use Chinese AI models

"These Chinese models are excellent," Huang told Axios on Tuesday. "Open-source models that are excellent should be used." He said companies should "absolutely" be allowed to use them. The remarks came as the release of Kimi K3, a model from Beijing-based Moonshot AI that combines near-frontier performance, lower prices, and open weights, has rattled chip and AI stocks, reviving concerns that cheaper models could undercut the case for large AI infrastructure spending. Huang argued Wall Street has the situation backward. "Free AI should be great for hardware," he said. "Free AI should be great for chips. Free AI should be great for data centers." Cheaper and more accessible models draw more people into AI ecosystems, he argued, and that broader adoption ultimately drives up demand for the chips and data centers that Nvidia provides.

As Washington panics about Chinese AI, Jensen Huang says open-source models like Kimi are 'excellent' and should be embraced, not banned | Fortune

Jensen Huang is urging Washington to stop worrying about China's open-source AI just as an all-out panic has broken out over the highly capable, low-cost models that can increasingly compete with the most advanced products by OpenAI and Anthropic. The Nvidia CEO said in an interview with Axios Tuesday the U.S. should not ban these kinds of models, which are being developed by Chinese companies like DeepSeek, Alibaba, and Moonshot AI -- which last week raised alarms when it released its highly capable model, Kimi K3. While U.S. companies, including AI coding startup Cursor, are increasingly turning to open-source alternatives from China to save on the high costs associated with American AI models, the White House has reportedly considered restricting their use partly based on fears that the models could be used for surveillance purposes by China or that they could threaten the viability of homegrown AI companies. The White House reportedly considered using executive power to impose conditions on U.S. companies looking to use Chinese AI models, including forcing these companies to guarantee security and accept liability if breached, Axios reported. The White House did not immediately respond to Fortune's request for comment. Still, Huang on Tuesday rejected the claim that these models could serve as a backdoor for China's government as a "misconception." These models are downloadable and the guardrails that can be put on them are customizable, he emphasized. He also said these open-source models aren't going to outcompete American AI. Rather, he argued, the world needs both open-source models and closed models, like those made by OpenAI and Anthropic. "These Chinese models are excellent," he said. "Open-source models that are excellent should be used." Nvidia did not immediately respond to Fortune's request for comment. Closed AI models like Anthropic's Fable 5 or OpenAI's GPT‑5.6 Sol, are controlled by the companies that create them. They usually offer users less freedom to inspect or modify how they work, but they can be easier to deploy because the developer operates and maintains the model. Open-source models, on the other hand, can be downloaded and adapted by companies for specific needs, but they also require that more time and money be invested into setting them up. Some companies prefer open-source models because they can run them on their own systems, giving them more control over sensitive data. With high usage, these models can be cheaper, but they also require companies, in some cases, to invest in powerful computers and ongoing maintenance. Closed models can be more expensive with higher usage, but they are more practical for companies that don't want to deal with the costs of setting up and maintaining the infrastructure to run them. Huang's comments on Chinese open-source models come after the release of Moonshot's Kimi K3 last week raised alarms in the U.S. The open-source model, released on July 16, is the company's most advanced open source model to-date and has been able to compete in some coding tests with Anthropic's most advanced, publicly-available model, Fable 5. Kimi K3 is also substantially cheaper than U.S. models, even though it is more expensive than other open source Chinese models. The model costs $15 per 1 million output tokens, while Anthropic's Fable 5 costs $50. Another popular Chinese model, DeepSeek V4, costs 87 cents for the same output, Fortune previously reported. As Washington panics, Huang insists allowing open-source models to be used by American companies would be good for both Nvidia and the broader industry. "If there's great AI, even if it's open, wherever it comes from, there will be more use," Huang said. "Whenever there's more use, you'll have to sell a lot more NVIDIA computers. We'll have to build more data centers. We'll have more services. The technology will diffuse into more industries..."

A Chinese AI Model Just Shot to Number One on the Charts, Sending Shockwaves Through the American Tech Industry

Can't-miss innovations from the bleeding edge of science and tech While Wall Street was fast asleep, a Chinese-made large language model quietly leapfrogged 16 other models to become number one on the AI charts. The model is called Kimi-K3, developed by Beijing-based firm Moonshot AI. On Thursday, the AI benchmark platform Arena.ai announced that Kimi had gone from number 17 in the "Frontend Code Arena" -- a measure of an LLM's ability to perform multi-step web development tasks -- to number one, surpassing the buzzy Claude Fable 5 and GPT-5.6 Sol by a mile. In the "Text Arena," a measure of an LLM's ability in text-to-text tasks like creative writing, Kimi-K3 earned the number nine spot, a significant improvement from Moonshot AI's previous model, Kimi-K2.6, which held number 38. The news comes as investors are facing a major reckoning, with US semiconductor stocks plummeting on Friday morning and the tech-heavy Nasdaq composite sliding by 1.4 percent. Those losses are extending a horrible week for tech stocks, which had been driven by concerns over an all-American AI bubble. The moniker, "Moonshot," might be an understatement. The major catch here is that not only did a Chinese AI model surpass every US-designed model in front-end coding in a benchmark, it did so using a dramatically different approach. Just like DeepSeek, a similar Chinese AI model that rankled the US stock market last year, Kimi is an open-weight model, meaning its inner workings are viewable to the public. Compared to proprietary models like GPT-5.6 that are kept under lock and key, open models cost users on average six times less, though their performance has historically been ever-so-slightly worse than their closed counterparts. Responding to the news, Xiaoyin Qu, former Meta senior product manager turned AI entrepreneur, posed an important question: "When the best open weight model exceeds the best closed-source model, how does [Anthropic] justify its Fable pricing? Why would anyone pay for that?" Even before Kimi-K3 dropped, the proposition of paying up to six times more for the slight performance boost offered by closed models was already pushing US companies toward Chinese AI. Now that the performance gap is closing fast, there's even less reason for companies or individuals to pay exorbitant prices associated with Silicon Valley's frontier models. That simple math is bad news for the US tech industry, which has spent years insisting that it will take trillions of dollars to make AI work. As Qu observed, "Kimi's most recent funding round values the company at $20 Billion as of two months ago. Anthropic is worth almost 1 trillion, 50x. Why?"

New open-weight AI from China is toppling the best of OpenAI and Claude Fable

China's Moonshot AI has launched Kimi K3, a massive 2.8-trillion-parameter model built for coding, research, reasoning, and visual tasks. Moonshot admits K3 still trails Claude Fable 5 and GPT 5.6 Sol overall. Even so, its benchmark results put it surprisingly close to both, and it finishes ahead in several tests. How close is Kimi K3 to the best closed models? K3 scored 77.8 on Program Bench, narrowly beating Fable 5 at 76.8 and GPT 5.6 Sol at 77.6. It also led BrowseComp with 91.2 and SWE Marathon with 42.0, ahead of both rivals. Recommended Videos Other results show there is still some distance to cover. K3 scored 67.5 on DeepSWE, compared to 70.0 for Fable 5 and 73.0 for GPT 5.6 Sol. Moonshot also says the overall user experience remains behind both proprietary models. These comparisons also come with an important caveat. Moonshot says all Fable 5 results may include fallbacks to another model, while GPT 5.6 Sol results may include cyberguards that restrict certain responses. In other words, the comparison is useful, but it is not perfectly even. Open weights put pressure on OpenAI and Anthropic The most important part of K3 may be its open-weight release. Moonshot plans to publish the full model weights by July 27, meaning you can download and run K3 locally, provided you have the hardware to handle a model this large. Developers will also be able to modify and fine-tune it for specific tasks. K3 is cheaper than the premium U.S. models it competes against, but it is not unusually affordable by Chinese AI standards. Its API pricing is $0.30 per million cached input tokens, $3 for uncached input, and $15 for output, putting it closer to Anthropic's mid-range models than to the heavily discounted prices associated with earlier Chinese releases. Chinese AI models are often far cheaper than their U.S. counterparts, and that gap is already pushing some American startups to adopt them to cut costs. Kimi K3 is not among the cheapest Chinese models, but its near-frontier performance at a lower price still puts pressure on OpenAI and Anthropic to justify their premiums.

Kimi K3 Just Triggered DeepSeek Flashbacks for the Stock Market

Full model weights go public by July 27. If they hold up under independent testing, the pressure on U.S. AI companies to justify their infrastructure spending gets significantly harder to dismiss. China did it again. Moonshot AI launched Kimi K3 in the dead of night and markets woke up today doing what they always do when a Chinese lab closes the gap: They panicked. Semiconductor and AI stocks dropped across the board Friday. Taiwan's benchmark fell more than 6%. Japan closed down 4%. The Nasdaq slid 1.5%, its worst session of the week. The DeepSeek comparison seems well deserved. When DeepSeek dropped R1 in January 2025, the assumption that frontier AI required frontier spending -- and the chip orders to match -- cracked overnight. Nvidia shed around $590 billion in market cap in a single session. This time the damage spread across the sector. The VanEck Semiconductor ETF (SMH) fell below its EMA support band -- the moving average tracking the price trend over the prior months -- for the first time since April, extending a rout that has put it more than 20% below its late-June record high. On the Artificial Analysis Intelligence Index -- an independent composite benchmark that aggregates model performance across reasoning, knowledge, mathematics, and coding -- K3 scored 57, ranking above Claude Opus 4.8 and GPT-5.5, practically on par with Claude Fable 5 and OpenAI's GPT-5.6 Sol, beating them in specific benchmarks at a fraction of the price. Full weights drop July 27 under a Modified MIT license which means small labs will have that model available for free. Wall Street analysts largely saw this coming. Bernstein's Robin Zhu called the release "confirmatory" -- another data point in a trend that's been building all year. Morgan Stanley analyst Gary Yu framed K3 as the product of steady compound progress rather than a shock. "K3 has received positive feedback globally, signaling an all-round catch-up of Chinese LLMs with U.S. leaders in model size, performance, and pricing," he wrote. Bernstein analyst Robin Zhu also framed this as a catch-up. "At a high level, K3 feels confirmatory of our views that (1) AI [state of the art] continues to evolve rapidly; and (2) China AI can continue to keep pace with global [state of the art], and take some share over time." Moonshot is backed by Alibaba, which put $1 billion into the company in 2024 at a $2.5 billion valuation. The startup now sits at roughly $31.5 billion. As Decrypt covered in May, Moonshot already has deeper roots in U.S. developer circles than most realize -- Cursor's Composer 2 was found to be running on Kimi K2.5 without disclosure before the Cursor team acknowledged the open-source base.

China's Kimi K3 rattles US AI industry

Washington (United States) (AFP) - Kimi K3, a new artificial intelligence program from a Chinese startup called Moonshot AI, stunned the US tech industry on Friday, setting off fresh discussion over the China-US rivalry to dominate AI. The program was released Thursday and within hours hit the top spot on a widely watched ranking of AI coding tools called Arena, marking the first time a Chinese model had claimed the number one position on the list. Investors in Silicon Valley and on Wall Street, as well as White House officials, worry that if China can build AI as good as America can, then US companies like OpenAI and Anthropic may struggle to keep charging high prices for their products. Kimi K3's release follows that of other much-hyped Chinese AI models. Like most of China's offerings, it costs less and uses source code that programmers can customize. Some experts compared the moment to early 2025, when another Chinese company called DeepSeek shocked markets by releasing a powerful AI model at a fraction of the usual cost. That episode briefly wiped hundreds of billions of dollars off the value of US tech companies. 'Reckoning' Anastasios Angelopoulos, who runs the Arena ranking site, told the TITV podcast that Kimi K3 could force investors to rethink the whole AI industry, since businesses may prefer free Chinese programs they can customize on their own computers over paid American ones that require sharing data with outside companies. He said the release will likely "cause a reckoning in the capital markets" since it "brings into question what the dominance will be" of US-based, closed-source models like those from OpenAI and Anthropic. David Sacks, a venture capitalist who advises the White House on AI and is a vocal opponent of tech regulation, said on X that Kimi's success showed US dominance was under threat and that the technology should be allowed to develop unimpeded. He argued that American politicians are slowing their country down by blocking new data centers, adding state-level rules and pushing for a federal agency to approve powerful AI models before they can be released. "This is how you lose the AI race," Sacks wrote. Dean Ball, who recently worked as an AI adviser in the White House and now works at OpenAI, said Kimi K3 was clearly a strong program and not just a copy of American models, a frequent criticism of Chinese AI products. He predicted President Donald Trump's administration will eventually try to more explicitly discourage US companies from using Chinese AI -- not by outright banning it, but by warning that it contains hidden risks and that companies should not use them. "You just create enough regulatory risk that every regulated enterprise backs off," he wrote on X. For Gavin Baker, a prominent Silicon Valley investor, Kimi K3 is "potentially negative for Anthropic and OpenAI while being net positive for essentially every other company in the world."

Trump's AI advisor says Kimi K3 shows "how you lose the AI race." Khosla blames immigration policy. Marcus wants a congressional investigation.

Kimi K3 topped Arena's coding leaderboard in 24 hours. Trump advisor Sacks warned the US is losing. Khosla blamed immigration. A Wall Street CIO called it bad news for closed AI labs. Moonshot AI's Kimi K3 topped Arena's frontend coding leaderboard within 24 hours of launch, ahead of Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 Sol. It placed third on Artificial Analysis's Intelligence Index. The reaction from Washington, Wall Street, and academia was swift and pointed. TNW covered the model's capabilities at launch, but the political and economic fallout is a story of its own. David Sacks, Trump's former AI czar and current cochair of the President's Council of Advisors on Science and Technology, called the release "concerning" and said it was the first time a Chinese model had taken the top coding spot. He blamed US regulatory self-harm: blocking data centres, layering on state regulations, and pushing for federal pre-approval of frontier models. "This is how you lose the AI race," Sacks wrote. "Permissionless innovation is how America won the internet." Vinod Khosla agreed and added that US immigration policy is "scaring away brilliant talent." Gary Marcus, the longtime AI industry critic, took a different view. "Congress should investigate. Seriously," he wrote, pointing to a Goldman Sachs chart showing US cloud capex reaching roughly $1 trillion in 2027, about eight times China's projected spending. The implication: if Chinese labs can match or approach US performance at a fraction of the cost, the trillion-dollar buildout may be harder to justify. Gavin Baker, CIO at Atreides Management, called K3 an "inflection point" and said it is bad news for closed AI startups like OpenAI and Anthropic but a net positive for everyone else. "Anything that lowers margins and increases competition at the model layer is good for every other AI layer: power, semiconductors, hyperscalers, neoclouds and yes even software," Baker wrote. DeepSeek already made its 75% price cut permanent, and K3's open-weight release at frontier-level coding performance accelerates the same dynamic: the model layer commoditises while the infrastructure and application layers capture the value.

Exclusive: Nvidia's Jensen Huang defends Chinese AI amid Kimi panic

Why it matters: The world's top AI chipmaker says American companies should "absolutely" be allowed to use Chinese models -- a direct challenge to Trump officials and the U.S. labs lobbying Washington to shut them out. Among them are Nvidia's own customers, OpenAI and Anthropic, which have accused Chinese rivals of siphoning capabilities from their models and warned that Beijing's open-source surge threatens America's AI lead. * Huang rejected the idea that OpenAI and Anthropic should be afraid of open models, arguing they expand the market by giving more people a first taste of AI, while many users will still pay for the convenience, reliability and stronger performance of closed services. * "There's no scenario where China runs U.S. companies off the road," Huang told us. "Zero possibility." * Huang sat down with Axios in Fort Worth, Texas, at the opening of a new phase of an advanced manufacturing plant where Taiwan-based Wistron makes AI infrastructure for Nvidia, showing the U.S. is capable of some of the world's most sophisticated manufacturing. The big picture: The release of Kimi K3 by Beijing-based Moonshot AI has triggered the most intense bout of AI panic since DeepSeek rattled financial markets in January 2025. * Kimi combines three qualities rarely seen together: near-frontier performance, dramatically lower prices and open weights that developers will soon be able to download and customize. * That combination helped fuel a sell-off of Nvidia and other chip stocks, reviving fears that cheaper, more efficient models could undercut the case for the industry's massive AI infrastructure buildout. Zoom in: Huang says Wall Street and Washington have the Kimi shock exactly backward. * "The market misunderstood the impact of DeepSeek the first time," he told Axios, adding that Wall Street has "misunderstood the impact of Kimi again this time." * His argument is simple: Cheaper, open models will bring AI to more people and businesses, increasing demand for the chips, data centers and computing power Nvidia sells. * "Free AI should be great for hardware," Huang said. "Free AI should be great for chips. Free AI should be great for data centers." Zoom out: When it comes to policy, Huang believes restricting open models in the name of national security could leave America more vulnerable. * He rejected the "misconception" that downloaded Chinese models create a "backdoor" to Beijing, arguing that companies can customize them and control their access inside secure "sandboxes." * Huang suggested openness makes AI more secure, not less, because outside researchers can inspect the models, expose weaknesses and build defenses. * "If everything just becomes one single model, one single point of attack, one single source of failure, I think the world is much, much more vulnerable," he said. Huang applied the same logic to American models, and called for Anthropic to make its restricted cyber model, Claude Mythos, available to "everyone" rather than limiting access. * Huang argued that companies should harden powerful models through testing and rapid fixes rather than restricting access from the outset. "Mythos should be available as a service," he said. "Remember: Just because Mythos is not available, open models are available anyhow. So I think: Let Anthropic run." * "Holding Anthropic back is not in the benefit of the United States," he said. Between the lines: Huang sees the AI race itself very differently from Washington, where some Trump advisers have cast China's gains as a five-alarm threat to U.S. dominance. * He dismissed the notion that it's "a race with an endpoint": "We're going to continue to use AI forever. The United States is going to be here for a long time. China's going to be here for a long time." The intrigue: Huang's comments came hours after Treasury Secretary Scott Bessent told Fox Business that the administration is examining Chinese AI models for stolen U.S. intellectual property -- and considering sanctions in response. * "If we see ... that overseas models are stealing from our great companies, we have the ability to sanction them because of this theft," Bessent said, citing "watermarks" of U.S. models inside Chinese ones. Huang drew a different conclusion. "Distillation, learning from AI, learning from other sources of knowledge, is fundamental to intelligence," he told Axios.

Is China replacing the U.S. as the world's AI touchpoint thanks to Kimi K3?

The release of Kimi K3, an artificial intelligence model created by a former computer science graduate student at Carnegie Mellon University, has rocked parts of social media and the AI commentariat. The model, developed by Moonshot AI -- a Chinese lab cofounded by that ex-student, Yang Zhilin -- has wowed many observers with its capabilities. "Kimi K3 demonstrates the U.S. moat in building frontier AI software is not as durable as many of us had hoped," says Ryan Fedasiuk, a fellow at the American Enterprise Institute. "We should expect Chinese AI labs to continue distilling and freely releasing a version of the American frontier at a pace just weeks behind U.S. labs." Social media has been ablaze with eye-popping demonstrations of the model's prowess, including its ability to spin up a convincing browser-based version of macOS in a matter of minutes. The demonstrations have led some observers to conclude that Chinese AI models have caught up with their U.S. counterparts. But Yang, and Moonshot itself, recognize what K3 is not. It is no Claude Fable or GPT-5.6 killer. The company's CEO even added a line to the model's release notes acknowledging a "noticeable gap" between K3 and the leading U.S. models.

Moonshot unveils Kimi K3, largest open-weight AI model yet

Moonshot announced a $2bn raise at $12bn valuation in May. China is showcasing its AI prowess despite efforts by Washington, as Moonshot AI's new Kimi K3 boasts a performance close to OpenAI and Anthropic's latest models. At 2.8trn parametres, Kimi K3 is the largest open-weight AI model available in the market today. The multimodal model is designed for frontier intelligence across long-horizon coding, knowledge work and reasoning. Across benchmarks, K3 only trails behind OpenAI's newest GPT-5.6 Sol and Anthropic's Fable 5, while outperforming them in some coding and general agent tasks. The company said that K3 performed "competitively" with Fable 5 and "substantially outperformed" Opus-4.8, GPT-5.6 Sol and GPT-5.5. K3 is also the cheapest of the three, at around $0.94 per Artificial Analysis intelligence index task and $15 per million output tokens, while GPT-5.6 Sol max costs around $1.04 and Claude Fable 5 makes a significant leap to around $2.75 per task. Although cheaper than the American juggernauts, Moonshot's new model marks a major leap in price compared to its Chinese contemporaries. DeepSeek's V4 Pro costs around $0.04 per index task according to Artificial Analysis rankings, while MiniMax's M3 costs around $0.12 for the same. Moonshot unveiled its new model just months after announcing a $2bn raise from Chinese food delivery company Meituan's VC arm Long-Z Investments, alongside Shuimu Capital, China Mobile and CPE Yuanfeng. The May round valued the start-up at around $20bn. The launch marks the latest in escalating tensions between US and China over AI leadership. US curbs on semiconductor exports to China over recent years has pushed China to focus on its own chip capabilities, with Moonshot AI president Yutong Zhang telling the audience at the World Economic Forum earlier this year: "We knew we didn't have the luxury to simply scale up compute...That forced us to focus on fundamental research and efficiency." The US is taking a more restrictive approach to sharing advanced home-grown AI models by only allowing approved bodies to access GPT-5.6, Fable and Mythos - much to OpenAI's disappointment. Lawmakers from the country, meanwhile, are also urging the Trump administration to ban US companies from buying memory chips from Chinese semiconductor developers. Don't miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic's digest of need-to-know sci-tech news.

China's Moonshot throws down the gauntlet with Kimi K3, the world's largest open-weights model

Chinese artificial intelligence lab Moonshot AI has just announced the imminent release of Kimi K3, its latest large language model. Already, it's sending shockwaves across the AI industry, not only because it's believed to be the world's largest open-source model to date, but because benchmarks show it outperforms the best models from OpenAI Group PBC and Anthropic PBC in some applications. With a staggering 2.8 trillion parameters, Kimi K3 will become the largest open-weight model available once its weights are released to the public on July 27. Moonshot, which is backed by Alibaba Group Holding Ltd., said in a blog post that Kimi K3's performance still trails GPT-5.6 Sol and Claude Fable 5 in some areas. However, the company's internal tests reveal that it's extremely close behind those models in several key tasks. Moreover, Artificial Analysis has already carried out a number of independent tests which show that it places just behind those top proprietary models on its Intelligence Index and in real-world work evaluations. Meanwhile, Arena.ai's front-end development leaderboard has Kimi K3 ranked above those models. That puts it 17 places above Moonshot's previous model release, Kimi K2.6. Arena Chief Executive Anastasios Angelopoulos made an extremely bold claim in a post on X, saying that it may be "the single biggest release of the year," and may represent the moment in which China has finally surpassed the U.S. in its AI model prowess. For an open-source model that is an exceptional achievement, and it means that the general consensus that American AI firms remain several months ahead of their Chinese counterparts no longer holds much water. Anthropic only released Fable 5 last month, while OpenAI's GPT-5.6 only debuted a week ago. Software engineering focus In a blog post, Moonshot explained that Kimi K3's primary use case is focused on long-running autonomous software development tasks. The model is designed to analyze large codebases, coordinate programming tools and perform multistep tasks in order to achieve an end goal. Kimi K3 also relies on visual feedback, the company explained. It's able to examine screen captures, modify code, then check the resulting visible output of its work. Moonshot said this is an example of a "vision-in-the-loop" system that makes it especially useful for tasks such as games development, user interface design and computer-aided design. A demonstration posted by Moonshot shows off a 3D open-world game that Kimi K3 reportedly built entirely in a web browser using Three.js, WebGPU and GPU Compute. For the demo, the model procedurally generated the environment and used external tools to create a 3D rider and horse. It also demonstrated a simulation of the Long March 10 rocket's launch and return, and a Game Boy Advanced emulator. The Kimi K3 API documentation shows that the model is priced at $0.30 per one million input tokens with a cache hit and $3.00 without. One million output tokens, including reasoning, cost $15. Those prices apply regardless of context length. That makes it significantly cheaper than Western models. For instance, Fable 5 costs $1 per one million input tokens and $50 per one million inputs, while GPT-5.6 Sol costs $0.50 per one million input tokens and $30 per one million output tokens. Another DeepSeek moment? Former White House policy advisor on AI Sriram Krishnan summed up how many people felt about the release, saying on X that Kimi K3's debut is a "big moment, with multiple implications for the entire industry." The last time that a Chinese AI lab released a cheaper model that was able to match the capabilities of its proprietary U.S. counterpart, it resulted in shockwaves that reverberated across the world. That was when DeepSeek Ltd. released its R1 model in January 2025, and the moment people became aware of just how good it was, it caused pandemonium on U.S. stock markets. Roughly $1 trillion was wiped from the value of leading technology firms, and its success sparked major security concerns in the White House, causing the Trump administration to double down on its hardline position regarding tech exports to China. Do not be surprised if accusations of foul play are later thrown at Moonshot. Just a few months earlier, Anthropic accused the company, along with DeepSeek and another Chinese model maker called MiniMax of violating its rules by extracting the capabilities of its Claude model to train their own. That process is known as model distillation, and it's actually quite common in the AI industry, though the Trump administration recently deemed it to be an "adversarial" approach and promised to crack down on it. The launch of Kimi K3 comes at a time of heightened scrutiny and growing security concerns over the capabilities of frontier models. The release could well lead to yet more debate over issues like distillation, export controls and whether the restrictions placed on China are having any real effect.

Moonshot AI Kimi K3 launch sends rival AI stocks lower

Moonshot AI launched Kimi K3, a 2.8-trillion-parameter model the Beijing-based startup says rivals leading systems from OpenAI and Anthropic, sending shares of Chinese AI competitors sharply lower on Friday. Kimi K3 is the largest open-weight AI model released to date, according to Moonshot. The model places below Anthropic's Claude Fable 5 and OpenAI's GPT 5.6 Sol in overall rankings, yet beat Claude Opus 4.8 and GPT 5.5 across coding and general agent evaluations. The model features a 1-million-token context window and native support for both text and images. The launch rattled markets. Z.ai shares lost as much as 30% of their value in Hong Kong trading -- the largest single-day decline the company has recorded since its January listing -- according to Bloomberg. MiniMax Group stock dropped as much as 16%. Alibaba stock fell 4%. Bloomberg's Asian semiconductor index slid more than 6%, and Nasdaq $NDAQ 100 futures fell 2%. Arena blind evaluations showed developers choosing Kimi K3 ahead of all top U.S. models on front-end coding tasks, and K3 matched GPT 5.6 Sol when ranked across Arena's general text category, according to Axios. Rather than undercutting the market, Moonshot has positioned the API near Anthropic's mid-range price tier, distancing K3 from the steep discounts that have defined most Chinese model releases, Axios reported. "K3 raises the capability ceiling for China AI models, shifting the burden of proof to other independent AI labs," Bank of America $BAC analyst Alex Liu said in a note, according to CNBC. Moonshot has set July 27 as the date for publishing K3's full weights, leaving developers without the ability to self-host or alter the model in the meantime. The company said the model is available now through Kimi.com, mobile apps, and its API. The release landed as Chinese President Xi Jinping made his first appearance at China's premier AI summit. Xi urged nations to work together on AI, declaring that the technology "should not be a solo performance by a single country." Kimi K3's launch follows OpenAI's broader rollout of GPT-5.6 Sol after the Trump administration approved general public access, ending weeks of restricted availability to government-vetted partners. Anthropic's most powerful model, Mythos 5, remains restricted to a narrow set of American organizations following a Commerce Department export control order. Moonshot was founded in 2023 and raised $2 billion at a valuation above $20 billion in May. Backers include Alibaba and Tencent.

A Chinese Pink Floyd fan is giving Claude and Chat their own DeepSeek moment -- an AI model just as good and half the price | Fortune

Another powerful new artificial intelligence model from China took the U.S. tech industry by surprise Friday, the latest sign that Chinese startups that publicly release their "open-source" AI technology are making the California titans of AI sweat. The newest Kimi K3 model from Beijing-based startup Moonshot, run by a Pink Floyd-loving entrepreneur who earned his doctorate in Pittsburgh, appears to be catching up to the best versions of Anthropic's Claude and OpenAI's ChatGPT. "This may be the single biggest release of the year," and marks a moment when open-source Chinese models are surpassing closed U.S. models, said Anastasios Angelopoulos, co-founder and CEO of Arena, a platform for evaluating AI systems. Kimi K3 topped the charts in Arena's ranking of what it calls "front-end coding capability," a measure of an AI large language model's performance. "More results are rolling in that are likely to continue to show it is at the top of the pack," Angelopoulos said on social media. It was not likely a coincidence that K3's unveiling came shortly before Chinese President Xi Jinping's opening address Friday to the nation's annual World Artificial Intelligence Conference in Shanghai. American-led restrictions have blocked China from accessing some of the world's most advanced technologies, spurring China's efforts to build its own know-how and intensifying the rivalry between the world's two biggest economies. "The development of artificial intelligence should not be a solo performance by any single country but rather a symphony of global cooperation," Xi said at the event. Chinese AI models have shown large strides K3 follows another major AI model release last month from the Chinese startup Zhipu, or Z.ai. Its new flagship GLM-5.2 model is already widely used by software developers around the world who say it can perform work almost as well as top U.S. models at a lower price. The hype over the new Chinese model resembles the market-shaking panic that followed Chinese startup DeepSeek 's new model release in early 2025, though not everyone finds it justified. The response to K3 is an "overreaction shockingly similar" to DeepSeek's release last year, said tech analyst Patrick Moorhead on social media. He said it could be good for parts of the broader AI industry but poses a revenue challenge to Anthropic and OpenAI. During the conference, which runs until Monday, tech giant Huawei has also been showcasing a new AI computing system called the Atlas 950 SuperPoD, a signal that China increasingly is amassing the domestic hardware it needs despite U.S. restrictions on imports from chipmakers like Nvidia. Moonshot hasn't said what hardware it used to build K3, but the startup is a partner with Huawei. The price to use K3 is the highest yet for a Chinese AI model, but is still half as expensive as OpenAI's high-performing GPT-5.6 Sol model, according to a Friday report by Bank of America research analysts. U.S. politicians and several major U.S. AI companies including Anthropic and OpenAI have accused Chinese AI models of illicit "distillation" of their models to extract their technologies, a claim that Beijing says is "groundless." Anthropic in February accused DeepSeek, Moonshot and a third China-based AI lab, MiniMax, of engaging in campaigns to "illicitly extract Claude's capabilities to improve their own models" using the distillation technique that "involves training a less capable model on the outputs of a stronger one." Anthropic said that distillation can be a legitimate way to train AI systems but it's a problem when competitors "use it to acquire powerful capabilities from other labs in a fraction of the time, and at a fraction of the cost, that it would take to develop them independently." But it can go both ways. San Francisco-based startup Anysphere, maker of the popular coding tool Cursor, has acknowledged that one of its top products was based on Moonshot's K2.5 model. Elon Musk's SpaceX is planning to close a deal to buy Cursor for $60 billion later this year. K3 marks a leap for 'open-source' AI models Moonshot co-founder and CEO Yang Zhilin earned his Ph.D. in 2019 at Carnegie Mellon University, where he is said to have made fundamental contributions to the machine-learning field and was known for a love of rock bands like Pink Floyd. The pride among his former colleagues at the Pennsylvania school transcends the U.S.-China rivalry. "What a huge win for the open-source community! It feels like just yesterday Zhilin was graduating from my lab at CMU," wrote his former adviser Russ Salakhutdinov, who is also a former director of AI research at Apple. Developers who build "open-source" AI make key components of the technology accessible for anyone to examine, modify and build upon. Proponents say open-source practices promote innovation, while critics warn that making powerful AI models publicly accessible poses safety and security dangers. ___ Associated Press writer Chan Ho-him contributed to this report.

China's Kimi K3 Is Out -- And Beats Claude Fable and GPT 5.6 Sol on Key Benchmarks

Full model weights -- the files that let anyone run, fine-tune, or build on the model locally -- drop by July 27 under a modified MIT license, making K3 the largest freely available AI model in history. Moonshot AI just put out the biggest Chinese open-source model ever released, and it topped Claude Fable 5 at writing scripts. Towards AI's Writing Elo -- a benchmark where models write real scripts judged blind against published versions, scored using the same Elo system that ranks chess players -- put Kimi K3 at 2,840, above Fable 5 (max) at 2,760. That's a ranking Anthropic's team has historically dominated. K3 also claimed the top spot on Arena AI's Frontend Code Leaderboard -- a ranking built from thousands of pairwise human votes on code generation tasks, again Elo-scored -- with 1,679 against Fable 5's 1,631. First place in six out of seven frontend domains. The Artificial Analysis Intelligence Index -- a score built from nine independent evaluations covering coding, reasoning, agentic work, and knowledge, rated 0 to 100 -- puts K3 at 57, with Claude Fable 5 at 60, GPT-5.6 Sol at 59, and Claude Opus 4.8 at 56. That places K3 as the third-most capable model on the composite, with Fable 5 beating it just by 3%. If you want to have an idea of what it can do, this is a zero-shot result of a prompt asking the model to build an iOS clone. For comparison, this is the best approximation shared on social media using GPT 5.6 Sol and a much elaborate prompt. What this thing actually is K3 packs 2.8 trillion parameters -- the numerical values that store a model's knowledge -- in a mixture-of-experts architecture. Mixture of experts splits those parameters into 896 "expert" subnetworks and activates only a fraction for any given task. That's how you get frontier-level intelligence without melting the server room. "It is the world's first open-source model in the 3-trillion-parameter class, designed for frontier intelligence scenarios including long-horizon coding, knowledge work, and reasoning," Moonshot AI says. That's not marketing theater: DeepSeek's V4-Pro tops out at 1.6 trillion parameters, Moonshot's own K2 at one trillion. K3 roughly doubles the nearest open-weight competitor on the size chart. It comes with a one-million-token context window -- tokens are the basic unit of information an AI processes, about three-quarters of a word each -- native image and video understanding, and always-on reasoning. Two architectural techniques underpin the efficiency gains. Kimi Delta Attention speeds up decoding for long sequences -- up to 6.3x faster at million-token contexts. Attention Residuals routes information selectively across model layers rather than accumulating it uniformly, adding about 25% training efficiency at under 2% extra compute cost -- together yielding roughly 2.5x better scaling efficiency than K2. Benchmarks are nice, Prices are nicer Kimi K3 costs $3 per million input tokens and $15 per million output tokens -- the same rate as Claude Sonnet 5, Anthropic's mid-tier model. The difference is that Sonnet 5 is Anthropic's middle-ground offering; K3 is sitting three points below Fable 5 on the Artificial Analysis composite. Per task across that nine-benchmark suite, K3 runs $0.94 versus $1.04 for GPT-5.6 Sol and $1.80 for Opus 4.8. In other words, this model offers top of the line performance at mid-tier level prices. As Decrypt covered in May, the pricing gap between Chinese and American frontier AI ran 15-30x earlier this year. K3 doesn't undercut at DeepSeek rates -- it prices like a Western mid-range model -- but delivers near-frontier performance at that level. For teams building on the API this represents a major cost improvement. If Anthropic goes ahead with its intentions of making Fable 5 available only via API, K3 becomes the nearest open-weight alternative to whatever model currently sits second in the industry -- at half the per-task cost of Opus 4.8. That's the scenario benchmark chasers are already running the math on. K3's launch is the argument U.S. chip export controls advocates don't want to have. The U.S. restricted Nvidia's H800 GPUs from export to China in late 2023; Moonshot confirmed it trained earlier models on those chips. K3's own benchmark documentation references H200s and what the company calls "a GPGPU from an alternative vendor" -- widely interpreted as Huawei Ascend hardware -- without specifying where that hardware sits. Moonshot AI president Yutong Zhang framed the constraint directly at Davos this year, per Silicon Republic: "We knew we didn't have the luxury to simply scale up compute... That forced us to focus on fundamental research and efficiency." Bank of America analysts, in a note after the launch, wrote that K3 proves "pre-training scaling, paired with architectural innovation, can still deliver step-change gains for flagship Chinese models" under those constraints. Moonshot is one of the so-called AI Tiger startups that have collectively shifted the global model landscape without access to the chips Washington said they'd need. Whether that's an argument for tighter export controls or an argument that they don't work is a policy question Washington hasn't settled. The asterisk you should read K3's hallucination rate on AA-Omniscience -- a benchmark that measures how often a model confidently fabricates an answer it doesn't know -- jumped from 39% to 51% compared to predecessor K2.6. More correct answers overall; more made-up ones too. The model also acknowledges in its own documentation that it can be "excessively proactive," making unexpected decisions on a user's behalf during long autonomous tasks. For teams that ran the Kimi K2.6-based tooling and want to upgrade, K3 is a meaningful step up on most fronts -- but that hallucination delta is worth stress-testing before you trust it with anything that needs to be accurate. If you want to try it for free, you can. It's available on Kimi's official website. But good luck: The servers are so packed that tasks get interrupted constantly due to traffic constraints, making it barely usable. A better alternative is to either pay for a subscription or use it over an API. Weights will be released on July 27. Those will be available for big enterprises and businesses. No domestic GPU, no matter how big, is currently able to handle a model this size.

Moonshot's Kimi upends conventional wisdom on US lead over China

The surprising gains in performance have stunned some AI watchers and investors, fuelling a tech rout with echoes of the DeepSeek moment last year. At an event in Beijing earlier this year, some of China's top artificial intelligence leaders warned that the country remained meaningfully behind the US in developing cutting-edge AI models, with one executive arguing that "the gap may actually be widening." Leading US firms also appeared confident they were significantly ahead. As recently as last week, one executive at Anthropic PBC, who spoke on condition of anonymity, mused that the Claude maker's technology was roughly six to 12 months ahead of Chinese rivals.

China's Moonshot AI chases 'DeepSeek moment' with much-hyped model

Shanghai (AFP) - A model released Friday by Chinese startup Moonshot AI has fuelled buzz around the country's tech prowess, as experts said it could rival some of the more advanced offerings from US labs. Large language models underpin chatbots and other artificial intelligence tools with their ability to crunch huge amounts of digital data. Moonshot AI's "Kimi K3" is one of several from China growing in global popularity thanks to their lower costs and source code that programmers can customise. Soon after its launch, Kimi K3 had topped a leaderboard for AI coding run by a platform called Arena created by UC Berkeley researchers. That drew excitement from industry insiders, with some evoking a 2025 release from China's DeepSeek that shook assumptions of US dominance in AI. "Kimi K3 seems really good, closest to the frontier yet," Ethan Mollick, a University of Pennsylvania professor and a leading voice on AI, said on X. But it "cannot write a good murder mystery (though neither can any other model). That remains the jaggedest of frontiers" of AI development, he said. "Sensing a violent market reaction to KimiK3... similar to DeepSeek moment," tech writer and investor Kevin Xu wrote. Beijng-based Moonshot AI said Kimi K3 was the world's first open-source model of its size. The more internal variables, or parameters, a model has, the better it can handle complex requests, and Kimi K3 has around 2.8 trillion. Leading US players Anthropic and OpenAI do not release details of how many parameters their top models have. 'Frontier-level' "Kimi K3 demonstrated frontier-level performance across our evaluation suite, consistently outperforming other tested models," Moonshot AI said. However, overall performance "still trails the most powerful proprietary models" from Anthropic and OpenAI, the company said. Hussein Abbass, a computing professor at UNSW Canberra, said Kimi K3 appears to be good at coding, "but it is still unknown how competitive it is across the whole range of tasks expected from foundation models". Should US rivals be concerned? "I wouldn't say they need to be worried. But they shouldn't be still," to maintain their edge, Abbass told AFP. He added that AI performance is "not just about the model" but also the hardware that runs it, along with data centres and supply chains. Arena ranked Kimi K3 ninth worldwide for text queries. When AFP tried Kimi K3, it generated ideas for tech reporting that compared favourably with responses from other chatbots. Kimi K3's release follows that of other much-hyped Chinese AI models such as Zhipu AI's GLM-5.2. It came during a major tech conference in Shanghai where President Xi Jinping on Friday urged international cooperation on AI governance. The United States, which restricts the export to its rival of powerful microchips that can train and run AI, said earlier this year it was around eight months ahead of China in the strategic field. But the Trump administration recently caused delays to the public release of top-end models from Anthropic and OpenAI, over concerns they could help hackers break into online systems. "Post-Kimi K3 and open weights models getting closer to the frontier again, I wonder if Anthropic and OpenAI will be allowed to increase their release cadence by the government," Mollick wrote.

AI race splits in two as China wages open-weight insurgency

Why it matters: Chinese labs like Moonshot AI are cornering the market for cheap, customizable intelligence, threatening to turn America's prestige models into expensive niche products. Driving the news: Companies were already shifting away from premium AI models and toward cheaper Chinese alternatives before Moonshot's Kimi K3 exploded onto the scene this week. * On OpenRouter -- a major marketplace that lets developers access hundreds of competing AI systems -- Chinese models now occupy the top five spots by weekly token usage. * All five models -- from China-based Tencent, Xiaomi, DeepSeek, MiniMax and Z.ai -- are "open-weight," allowing users to download, customize and run them on their own systems. Between the lines: Most corporate AI work does not require the smartest model available. * Businesses can use cheaper systems for routine coding, summarization, data extraction and customer service, reserving premium models for their hardest problems. * "There are going to be open‑source models that eventually handle 95% of enterprise queries, and that remaining 5% may go to OpenAI or Anthropic," one AI investor told Axios. What they're saying: Kong CEO Augusto Marietti told Axios that open-weight use has surged over the past quarter since flagship models are "too expensive." * Mozilla CTO Raffi Krikorian compared using frontier AI for everyday work to "driving a Ferrari to Whole Foods." * For many routine tasks, he said, cheaper models are fast enough, capable enough and can cost up to 50 times less. Threat level: Chinese models are advancing at astonishing speed, just as their lower prices and open weights make them easier to adopt. * Anthropic CEO Dario Amodei said in May that China remained six to 12 months behind the U.S. in the most dangerous cyber capabilities. * Ten weeks later, Moonshot released a model that rivals Anthropic's Fable and OpenAI's GPT-5.6 in key benchmarks -- underscoring how quickly any American lead can shrink. The other side: Some American companies are scrambling to fight back as China threatens to run away with open-weight AI. * Thinking Machines, a startup launched by former OpenAI CTO Mira Murati, made its highly anticipated debut this week with an open-weight model built for deep customization. * Nvidia is rapidly expanding its Nemotron family of open models, betting that customizable AI will drive more demand for the company's chips and software. * SpaceXAI this week open-sourced Grok Build, the software behind its coding agent -- extending the push for openness beyond the models themselves. The big picture: If businesses can get most of what they need from cheaper models they control, Silicon Valley's crown jewels may struggle to justify their premium prices -- and the enormous investments behind them. * OpenAI and Anthropic are preparing for blockbuster IPOs whose valuations depend on frontier AI remaining scarce, indispensable and lucrative. * Any rupture would reverberate far beyond Silicon Valley: AI spending is carrying an outsized share of U.S. growth, and the stock market has become highly dependent on a small group of companies riding the boom. The bottom line: "They're clearly terrified," Mozilla's Krikorian said of the U.S. labs confronting the rapid rise of Chinese competitors.

Chinese model Kimi 3 adds pressure on Trump administration's AI policy

The release of the Chinese AI model Kimi 3 is forcing Washington and Silicon Valley to reckon with another sign of intense competition with Beijing, renewing questions over how to curb the country's technology capabilities. Kimi 3, released late last week by the China-based firm Moonshot AI, quickly sent shockwaves through industry and policy circles, threatening the leads in coding ability of U.S. labs and models like Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 Sol. The model's release is boosting pressure on the Trump administration to keep the country's edge over China as it tries to pin down its domestic AI policy amid intense political backlash and intraparty divisions. Chris Lehane, chief global affairs officer at OpenAI, told reporters on Tuesday that he thinks the introduction of Kimi 3 "only reinforces ... the importance of the U.S. really establishing a coherent framework and process" for AI model testing. "There is a benefit of having a really clearly defined process so that you could speed the market and have real clarity on being able to get this out to various defenders," Lehane added, referring to government and critical infrastructure. President Trump signed an executive order in early June laying out the process for a voluntary testing framework, in which artificial intelligence companies can share their models with the government for up to 30 days before releasing them publicly. It gave agencies 60 days, or until Aug. 1, to create a classified benchmarking process for "covered frontier" AI models and a voluntary framework for companies to abide by. The White House's stance on AI development has fluctuated in the meantime, leaving companies in limbo. Anthropic and OpenAI stalled their latest model rollouts to the public, at the request of the government last month. David Sacks, former White House AI and crypto czar, argued last week the U.S. is risking its competitive edge. "Meanwhile America is tying itself in knots: politicians and bureaucrats are banning new data centers, piling on state regulations, and pushing for new federal agencies to pre-approve frontier models," the venture capitalist wrote on social media. Concerns over AI IP theft Tensions between the U.S. and China over AI development have escalated in recent months, fueled by allegations of intellectual property (IP) theft from Chinese companies, who are suspected of stealing American technology to build their own platforms at a lower cost. Some in the tech and policy communities were quick to speculate this was the case with Moonshot AI and Kimi 3, while others downplayed the role. Hamish Low, a research associate in compute policy at the Institute for AI Policy and Strategy, reviewed Kimi 3 and told The Hill that it's "pretty overwhelming likely they are distilling from leading U.S. models." He pointed to how Kimi is excelling in coding but not necessarily in cybersecurity. This is "pretty much what we would expect if they were doing lots of distillation of coding data, which they can more easily access from these leading models," Low said, adding Kimi is not on par with U.S. companies, which are more heavily "safeguarded" by the U.S. firms. Distillation occurs when the outputs of a stronger model are used to train less capable models. It comes about six months after Anthropic went public with concerns about Moonshot -- as well as other Chinese-based firms such as DeepSeek and MiniMax -- and accused them of distilling frontier models from U.S. companies for AI development in Beijing. The AI firm said at the time that, combined, the companies created more than 16 million interactions with Claude using about 24,000 fake accounts. Screenshots of users' purported conversations with Kimi 3 circulated online over the weekend showing the model answered, "I'm Claude, an AI assistant made by Anthropic," when asked for its name. Treasury Secretary Scott Bessent on Tuesday warned the Trump administration could sanction China over IP theft, while clarifying the government's stance on open-source models. "We've seen a lot of talk about ... open-source models coming and threatening the large language models in the U.S.," Bessent told Fox Business's Maria Bartiromo. "What I will tell you is this administration supports open-source models, but what we do not support is IP theft," he continued. "And if we see, especially that overseas models are stealing from our great companies, we have the ability to sanction them because of this theft." Meanwhile, OpenAI executive Dean Ball wrote on social media that he doesn't think Kimi's performance "can be explained away by distillation or anything like that." Pressed further on this, Ball added distillation "plays a role" but "clearly it's not primary." The Trump administration repeatedly called out Chinese-based firms earlier this year, accusing them of distilling frontier models from U.S. companies for AI development in Beijing. Michael Kratsios, the director of the White House Office of Science and Technology, wrote in an April memo the U.S. government is aware Chinese entities allegedly leveraged thousands of proxy accounts and jailbreaking tactics to access proprietary information. Some policy analysts suggested Kimi's release should be a wake-up call to do more to curb Chinese competition. "This is showing the lack of action from the administration to do anything in the China AI space, whether it's cut them off from U.S. technology in any way, or even prohibit certain uses [and sales] of Chinese models domestically," said Chris McGuire, who led U.S.-China AI policy at the National Security Council during the Biden administration. The lack of action, according to McGuire, is "coming home to roost, and it's presenting more of a strategic problem for the U.S." Debate over Chinese, open-source models Unlike its American competitors, Kimi will be released next week as an open-weight model, meaning the ways a model is trained to sift through information and formulate answers are made public. The structure gives individuals, companies and governments the ability to modify the model to their specific needs. The Trump administration reportedly weighed whether to ban the U.S.'s use of Chinese models -- which are often open-source -- in the wake of these models narrowing the U.S. lead on Beijing. A White House official said the administration is "committed to promoting America's open-source ecosystem and strengthening its security," but did not specifically mention Chinese models. It comes amid a wider debate around open-source AI and its role in U.S. technology development as private firms like OpenAI and Anthropic surge in pricing and find themselves vulnerable to the Trump administration's policy changes. Unlike private models, open-source models live in the public domain where any individual or business can download and customize them for personal use. These systems can be used, modified, examined and shared with anyone, for any purpose, and can be cheaper than private models offered by Anthropic, OpenAI and others. Kimi's release spurred debate among tech developers and policy analysts over how to approach China's latest artificial intelligence breakthrough. Ball, who also co-authored Trump's AI Action Plan, was slammed over the weekend for suggesting Kimi could prompt the Trump administration to look into creating more "regulatory risk" around the use of open-weight Chinese models. Sacks questioned whether Ball was suggesting "regulatory capture," and emphasized how Silicon Valley "still values open competition" with China. According to research from venture capital firm Andreessen Horowitz, 80 percent of developers around the world who use open-source tools are building with Chinese models due to the rising costs of private large-language models. This includes large companies in the U.S. Last year, Airbnb CEO Brian Chesky said the company was "relying a lot" on Alibaba's Qwen model and described it as "very good" and "fast and cheap." "What Kimi 3 really underscored is that there is a time frame between where we are and then where some of these other models may be," Lehane added. "There's a debate about whether or not [China] is closing the gap or staying steady, but either way, they're not falling further behind," McGuire said, adding Kimi makes this point "very clear."

Why China's New AI Model, Kimi K3, Has Silicon Valley and Washington on Edge

Chinese AI startup Moonshot on Friday unveiled Kimi K3, an open-weight model, meaning developers can access, study, and build on top of its architecture and users can download, run and customize the underlying systems, unlike closed-source models like the ones most commonly used in the U.S.. Moonshot claims Kimi K3 surpasses top systems from both OpenAI and Anthropic in some benchmarks (though it falls short of Claude Fable 5 and GPT 5.6 Sol on overall performance). Kimi K3 is the largest AI model to come from China to date, with 2.8 trillion parameters, a figure that refers to the size of its neural network. And AI experts are saying it could signal a tipping point in the industry.

Kimi K3 spooked markets. The AI selloff was already loaded.

Moonshot's Kimi K3, a 2.8-trillion-parameter open model billed as the largest ever, helped trigger a global tech and semiconductor selloff, reviving "DeepSeek moment" comparisons. The angle: the rout was multi-causal (weak Netflix and TSMC earnings, the Iran war, risk-off, rate fears), with Kimi one trigger among several. The deeper fear it amplified is real: if capable AI is becoming free, the ~$700bn hyperscalers are spending may not pay back, a risk Apollo's Torsten Sloek warned could tip the economy into recession. A new Chinese AI model has rattled global markets, sending tech and chip stocks sharply lower. Moonshot's Kimi K3 surprised investors and helped fuel a broad selloff, Bloomberg reports. Traders quickly labelled it a "Kimi moment", echoing the DeepSeek shock of early 2025. The comparison is doing a lot of work, and not all of it is fair. What Kimi K3 is By size, Kimi K3 is the largest open AI model yet released. It is a 2.8-trillion-parameter system with a million-token context window, and TNW has covered its debut as the world's largest open model. The claims behind it are what unsettled investors. Moonshot says it matches or beats the strongest US models, including GPT-5.6 and Claude Fable 5, on coding and long-horizon tasks. Crucially, it is open. The full weights are due later this month, so anyone can download and self-host a near-frontier model rather than paying OpenAI or Anthropic for access. The rout had many authors Here is where the framing needs a caveat. Friday's selloff was overdetermined, and Kimi was one trigger among several. Disappointing earnings did much of the damage, with Netflix down around 9% and TSMC lower after its results. Add the Iran war, a broad risk-off mood, and lingering rate and inflation fears. The moves were global and steep. South Korea's KOSPI fell more than 6%, Japan's Nikkei over 4%, and US chipmakers Intel, Micron, AMD, and Marvell all slid. The fear underneath Strip away the noise and one worry connects it all. If capable AI is becoming cheap or free, the hundreds of billions being spent to build it may not pay back. Apollo's Torsten Sloek warned of exactly this, that a timing mismatch between hyperscaler capex and revenue could tip the economy into recession if price competition from Chinese and open models undercuts AI income. Hyperscalers are on course to spend around $700bn on AI infrastructure this year. A free Chinese model that rivals the best paid ones is a direct challenge to that arithmetic. Why the panic may be overdone The DeepSeek precedent cuts both ways. That selloff was also billed as the end of the US AI trade, and the market recovered as the capex kept flowing. A model existing is not the same as enterprises adopting it. Trust, support, security, and integration still favour the incumbents, and US AI firms are, unlike the dot-com era, genuinely and heavily profitable. Even so, the pattern is hard to ignore. Chinese labs keep shipping cheap, open, near-frontier models, from DeepSeek, which just cut its prices by 75%, to MiniMax's giant open system. The real signal The takeaway is not that Kimi K3 broke the market. It is that markets are now primed to sell first and ask questions later whenever China shows the frontier can be reached on the cheap. Each of these launches erodes the same assumption, that frontier AI has to be expensive and American. That is the anxiety no earnings beat quite settles, and Kimi K3 has just poked it again.

Markets may have just experienced their second DeepSeek shock, this time thanks to a Chinese AI lab named after a Pink Floyd album | Fortune

On Thursday, Moonshot AI, a Beijing-based AI startup, unveiled the latest version of its Kimi large language model, which promises performance close to Anthropic's Fable 5 -- perhaps the most powerful publicly available model today -- at a fraction of the cost. Kimi K3, the largest open-weight model ever released, performed "competitively" with Fable 5, and "substantially outperformed" Anthropic's Opus 4.8, and OpenAI's GPT 5.6 Sol, according to Moonshot. The company's officially-released benchmarks consistently rank K3 among the top three AI models; one independent benchmark from Arena.AI even put K3 as the best model currently available, ahead of Anthropic. Many observers, including Anthropic CEO Dario Amodei, didn't expect a Chinese AI lab to release a model that could approach the U.S.'s best offerings for another six months. Tesla CEO Elon Musk, for example, suggested it might happen by the first quarter of next year. K3's launch rapidly shrank that timeline and underscores just how quickly Chinese AI developers are closing the performance gap with their U.S. rivals. "The AI ecosystem in China is probably much better than people thought," says Paul Triolo, a partner at DGA-Albright Stonebridge Group. Moonshot AI has priced K3 at $15 per million output tokens, the bits of information processed and generated by a large language model. That's cheaper than Fable 5, which costs $50 for the same level of output. Moonshot's launch seemed to rattle investors who interpret the new model as undermining the conventional wisdom that U.S. firms can maintain their extended lead by simply outspending Chinese competitors on computing power. A semiconductor selloff was already underway, but the K3 debut seemed to make things worse. Leading chipmaker Taiwan Semiconductor Manufacturing Company fell by 7% on Friday, despite reporting a 77% jump in quarterly operating profit. SoftBank -- often seen as a proxy for OpenAI -- fell by 9.0%. Z.ai, a Chinese AI startup that has released a competing model to Moonshot's Kimi, plunged by almost 30% in Hong Kong trading. Fear has spread into U.S. markets, with the Nasdaq 100 down by 1.0% as of 2:00pm Eastern time. Nvidia shares fell by 1.2%, briefly forfeiting the chipmaker's lead as the world's most valuable company to Apple. Meta shares plunged by over 2.4%. Who's who in China's AI sector China's AI sector includes both established tech companies, like Alibaba and ByteDance; smaller startups like DeepSeek, MiniMax, and z.ai; and even dark horse entrants like phonemaker Xiaomi and food delivery platform Meituan. Moonshot AI is one of the upstarts. Founded in 2023, its Chinese name comes from Pink Floyd's album The Dark Side of the Moon, which is founder Yang Zhilin's favorite album. The company is backed by Alibaba, Tencent, and Meituan, as well as HSG (formerly Sequoia China); it's reportedly considering an IPO in Hong Kong. This is Moonshot's second time in the spotlight this year. Earlier Kimi models, notably K2.5 and K2.6, gained traction among Silicon Valley developers by offering strong coding performance at meaningfully lower cost than Anthropic's Claude. In March, U.S. coding assistant maker Cursor acknowledged that its Composer 2 agent ran on top of Kimi 2.5. It's now taking its turn as China's AI leader, a position that's changed hands repeatedly this year. In April, DeepSeek unveiled V4, its long-awaited update to last year's V3 model. While V4 didn't shock markets quite as much as its predecessor, it still offered frontier-level performance at rock-bottom prices -- currently $0.87 for a million tokens of output -- and its ability to be run on Huawei-made processors. Fellow AI startup z.AI followed up in mid-June, with the release of its GLM-5.2 model. The Beijing-based startup announced the model just days after U.S. officials briefly disrupted access to Anthropic's Fable and Mythos models for some users outside the United States and for foreign nationals. "Frontier intelligence should not belong to only a few people, nor be subject to withdrawal by a handful of rules at any moment," the company wrote in a social media post announcing GLM‑5.2. Even China's consumer‑internet giants are getting into the frontier race. Meituan launched its own LongCat 2.0 model last month and revealed that it trained the system entirely on Chinese‑made semiconductors rather than U.S. chips. "The idea that Meituan could train a 1.6 trillion-parameter model on domestic hardware would have been inconceivable in October 2022," Triolo says, referring to the month when the U.S. launched sweeping export controls designed to slow China's access to advanced AI chips. Why are Chinese models cheaper? Chinese AI has earned a global following by being systematically cheaper than the U.S. competition. DoorDash, for example, is pushing "lower-level work" to Moonshot's Kimi model, leading to "better quality, cheaper cost," according to chief technology officer Andy Fang. Chinese systems routinely dominate the weekly leaderboards on OpenRouter, a popular marketplace that lets developers mix and match models from different providers. As of this week, all of the top five most-used models are from Chinese companies: Tencent, Xiaomi, DeepSeek, MiniMax, and z.ai. Power costs in China are lower than in many parts of the U.S., in part because the country has made massive investments into power generation and transmission, making it easier to add data‑center capacity. New data centers in the U.S., however, often face political resistance due to perceived strain on grids and water use. Chinese AI companies are also willing to sacrifice profit margins in a bid to capture market share and establish their models as a de facto standard. Ironically, U.S. export controls -- designed to hold China's AI sector back -- may also be a factor behind the low price of Chinese AI models. Without access to the most powerful AI processors, domestic labs were forced to squeeze more performance out of less capable hardware. "Labs are so compute-constrained, capital-constrained, and talent-constrained that a lot of them are being cautious in how they use their resources," says Grace Shao, an AI analyst and author of the AI Proem newsletter. Most recent Chinese AI models are also now compatible with cheaper, locally-made processors. "For the money [a Chinese AI company would] spend on an Nvidia chip, they can buy ten local chips from Huawei or other local chipmakers," says George Chen, a partner at the Asia Group. Perhaps most importantly, Chinese companies have embraced the open-source movement, releasing their models for free. Almost all Chinese companies release their work under licenses that allow users to download models for free, run them on local hardware, and fine-tune them. This allows third-party providers, including those in the U.S., to host Chinese models for free. The only costs that matter, in that case, are "GPUs and energy," says Ameya Karnitkar, co-founder of Larridin, an AI measurement platform. "Anthropic adds the additional costs of their R&D, and their own inference costs." On July 17, at the World Artificial Intelligence Conference in Shanghai, China President Xi Jinping affirmed that the country will remain committed to releasing models on an open-source basis. "We must seize this rare historic opportunity, encourage open source, openness, cooperation, and sharing, and comprehensively promote AI technological innovation, industrial development, and scenario-based applications," Xi told conference attendees. He also added a dig against Washington's attempt to slap controls on the release of frontier models. "We should jointly oppose the practice of overstretching the concept of national security in the field of AI or placing one country's own security above the security of others."

China Just Released Two AI Models That Claim to Rival OpenAI and Anthropic. And They're Free.

China landed a one-two punch on America's AI lead, and neither blow will cost users a dime. On Friday, The Verge reports that Moonshot AI unveiled Kimi K3, a 2.8 trillion-parameter model that the company claims ranks above nearly every U.S. system except OpenAI's GPT-5.6 Sol and Anthropic's Claude Fable 5. Days later, Alibaba followed with a preview of Qwen3.8, a 2.4 trillion-parameter model it calls "second only to Fable 5." The bigger story is the price tag. Both companies are releasing their models publicly, free for developers to download, modify and build on. Most US labs keep their best models locked up. China is giving theirs away. It's the biggest jolt to the industry since DeepSeek's low-cost model rattled Silicon Valley last year, and it raises a real question for American AI companies: does pouring billions into chips and data centers still guarantee they'll win the AI race?

China's Kimi K3 rattles US AI industry

Kimi K3, a new artificial intelligence program from a Chinese startup called Moonshot AI, stunned the US tech industry on Friday, setting off fresh discussion over the China-US rivalry to dominate AI. The program was released Thursday and within hours hit the top spot on a widely watched ranking of AI coding tools called Arena, marking the first time a Chinese model had claimed the number one position on the list. Kimi K3, a new artificial intelligence program from a Chinese startup called Moonshot AI, stunned the US tech industry on Friday, setting off fresh discussion over the China-US rivalry to dominate AI. The program was released Thursday and within hours hit the top spot on a widely watched ranking of AI coding tools called Arena, marking the first time a Chinese model had claimed the number one position on the list. Investors in Silicon Valley and on Wall Street, as well as White House officials, worry that if China can build AI as good as America can, then US companies like OpenAI and Anthropic may struggle to keep charging high prices for their products. Kimi K3's release follows that of other much-hyped Chinese AI models. Like most of China's offerings, it costs less and uses source code that programmers can customize. Some experts compared the moment to early 2025, when another Chinese company called DeepSeek shocked markets by releasing a powerful AI model at a fraction of the usual cost. That episode briefly wiped hundreds of billions of dollars off the value of US tech companies. 'Reckoning' Anastasios Angelopoulos, who runs the Arena ranking site, told the TITV podcast that Kimi K3 could force investors to rethink the whole AI industry, since businesses may prefer free Chinese programs they can customize on their own computers over paid American ones that require sharing data with outside companies. He said the release will likely "cause a reckoning in the capital markets" since it "brings into question what the dominance will be" of US-based, closed-source models like those from OpenAI and Anthropic. David Sacks, a venture capitalist who advises the White House on AI and is a vocal opponent of tech regulation, said on X that Kimi's success showed US dominance was under threat and that the technology should be allowed to develop unimpeded. He argued that American politicians are slowing their country down by blocking new data centers, adding state-level rules and pushing for a federal agency to approve powerful AI models before they can be released. "This is how you lose the AI race," Sacks wrote. Dean Ball, who recently worked as an AI adviser in the White House and now works at OpenAI, said Kimi K3 was clearly a strong program and not just a copy of American models, a frequent criticism of Chinese AI products. He predicted President Donald Trump's administration will eventually try to more explicitly discourage US companies from using Chinese AI -- not by outright banning it, but by warning that it contains hidden risks and that companies should not use them. "You just create enough regulatory risk that every regulated enterprise backs off," he wrote on X. For Gavin Baker, a prominent Silicon Valley investor, Kimi K3 is "potentially negative for Anthropic and OpenAI while being net positive for essentially every other company in the world.

Sacks argues US is 'tying itself in knots' over AI after release of new Chinese model

Former White House AI and cryptocurrency czar David Sacks argued Friday that the U.S. is "tying itself in knots" over AI and risking its competitive edge, following the release of a new Chinese model. Kimi K3, a new model from the Chinese startup Moonshot AI, sent shockwaves through the American AI industry Thursday after its developers claimed it could perform on par with Anthropic and OpenAI's leading models. The AI evaluator Arena also bumped Kimi to the top of its rankings for front-end coding, prompting Sacks to describe the development as "concerning." "Meanwhile America is tying itself in knots: politicians and bureaucrats are banning new data centers, piling on state regulations, and pushing for new federal agencies to pre-approve frontier models," the venture capitalist wrote in a post on X. "This is how you lose the AI race," he continued. "The rest of the world won't play by our rules if we bog ourselves down. Permissionless innovation is how America won the internet and became the technological envy of the world. We can do it again with AI -- while addressing risks in a targeted way -- or we'll watch our lead evaporate." Sacks, who announced his departure from the White House in late March, notably appeared to take aim at the Trump administration's recent approach to AI regulation. In the face of new, more powerful models, the White House has struggled to balance its avowed commitment to unhampered innovation with growing cybersecurity concerns. President Trump signed an order in early June to create a voluntary testing framework in which AI companies could share their models with the government up to 30 days before releasing them publicly. It marked a scaled-back version of an order that Sacks reportedly opposed. However, following the release of Anthropic's new Fable and Mythos models last month, the administration hit the company with an export control order, prompting the firm to pull both models. The White House ultimately lifted the restrictions in late June. OpenAI separately announced it would first roll out its new GPT-5.6 model to a limited group of partners at the request of the government. It publicly released the model several weeks later in July. The White House emphasized at the time that it did not give the company a "green light" approval to release its models and that "no such permission is required or granted."

Kimi K3 shocked the world. These other AI models could be next

Why it matters: The advent of AI has led to a global competition for supremacy, perhaps with the highest stakes since the development of the nuclear bomb. Driving the news: Kimi K3, a massive new model by Beijing-based Moonshot AI, all but erased the United States' lead in the AI race. * Kimi soared into the top tier of global AI, climbing over Anthropic's Fable 5 and OpenAI's GPT-5.6 Sol in front-end coding tests by AI evaluator Arena. Friction point: A report from the Five Eyes intelligence alliance in April warned that frontier AI capable of crippling governments and businesses is close. * Some of those models may have fewer guardrails than those imposed by the U.S. on the likes of OpenAI and Anthropic. * "The timeline is not years, it is months," Five Eyes warned. Here's what to know about top AI models overseas. China's other prominent AI models China's prominent LLM models -- other than Kimi K3 -- have recently closed the gap with Western countries. * DeepSeek has corralled billions in funding as a direct competitor to U.S. models. DeepSeek V4 rocked the industry in 2025. * Qwen3 has worried U.S. and European regulators given concerns about parent Alibaba's alleged ties to state entities, which the company has denied. * GLM-5.2 has a 1 million token context window and has been excelling in agentic work. The intrigue: Anthropic accused China's DeepSeek, MiniMax and Moonshot of using 24,000 fraudulent accounts with over 16 million exchanges "to illicitly extract Claude's capabilities to improve their own models," per Anthropic -- a process known as distillation. Yes, but: Dean Ball, OpenAI's head of strategic futures, says Kimi is so good that he doesn't think "its performance can be explained away by distillation or anything like that." Japan AI models to know Japan has a slew of growing AI models. * PLaMo is one of the most respected Japanese AI models. Its parent company, Preferred Networks, has been deploying specialized versions for finance, healthcare, and manufacturing, which could impact the U.S. hold on building AI for businesses. * Sakana comes from former Google researchers. Japan's Sakana AI launched Fugu Ultra, which the company claims can hit Mythos-level performance by using U.S. labs' work as interchangeable infrastructure. * Rakuten is mainly optimized for Japanese speakers. It has a foothold already in the ecommerce and banking space. Europe's Mistral Mistral represents Europe's best answer to OpenAI, Anthropic and Chinese AI giants. * The France-based company has built open-weight models for coding, reasoning, images and autonomous tasks, which gives European enterprises a way to use AI without handing over their data to American companies. Canada's Cohere The Toronto-based Cohere has become a direct competitor of OpenAI, Anthropic, Google and Microsoft in the business space with its Command A+ models. * Cohere's Command A+ models focus on offering private deployable systems that allow companies to keep their sensitive internal data in-house. Its recent deal with Germany's Aleph Alpha only accelerated the push for European sovereignty from American companies. South Korea, UAE models Zoom in: Both South Korea and the United Arab Emirates have top-tier models that are reshaping the space within their countries. * Naver's HyperCLOVA X is optimized for the Korean market and can be customized with a customer's business data. * United Arab Emirates' Falcon family of models focus on reasoning and deployment on smaller devices. The model primarily was developed around the Arabic language and for regional use cases, and offers Middle Eastern governments an alternative to American and Chinese prodcuts. Go deeper: Global AI wars

Moonshot unveils Kimi K3, the world's largest open AI model

The Chinese startup says its 2.8-trillion-parameter model outguns every open rival. The weights that would prove it are still two weeks away. Chinese startup Moonshot AI has unveiled Kimi K3, a 2.8-trillion-parameter system it bills as the world's largest open-weight AI model. The claim, made on 16 July, plants the firm squarely alongside the American frontier labs it has spent two years chasing. The model arrives weeks after Moonshot was reported to be seeking a $30bn valuation, and it reads like a pitch to justify the figure. On the company's own benchmarks, K3 ranks second overall behind only Anthropic's Fable 5 and OpenAI's GPT-5.6 Sol, the two closed models it is trying to catch. Moonshot is best known outside China for Kimi, the consumer chat assistant that gave the startup a following among developers before it began releasing frontier-scale weights. K3 is the point at which that reputation meets the biggest model the company has shipped. The system is a sparse mixture-of-experts design, activating roughly 50 billion of its 2.8 trillion parameters for any given token by routing through 16 of 896 experts. It carries a 1-million-token context window and ships with what Moonshot calls Kimi Delta Attention, a mechanism the firm says decodes up to 6.3 times faster over million-token inputs. The architecture is where the company plants its flag. Moonshot claims K3 achieves roughly 2.5 times better scaling efficiency than last year's Kimi K2, helped by a second trick it labels Attention Residuals, which it credits with about 25% higher training efficiency at under 2% extra cost. The "largest open model" billing is Moonshot's, and for now it is hard to dispute. DeepSeek's V4-Pro tops out at 1.6 trillion parameters and Moonshot's own K2 at 1 trillion, so K3 roughly doubles the nearest open competitor. Grok 4.5, by comparison, is reckoned to sit near 1.5 trillion. The catch is timing. Moonshot has published specifications and scores but will not release the weights until 27 July, which means no outside researcher can yet confirm the parameter count or reproduce the benchmarks. That gap between announcement and download matters more than usual here. The figures that make K3 impressive, the size and the scores alike, are for now the company's own, and early leaderboard results tend to flatter models before independent testing catches up. Where independent testing does exist, K3 looks strong rather than untouchable. It scored 77.8 on Program Bench and 93.5 on GPQA-Diamond, leading the American pair on several coding suites, but trailing them on harder software-engineering tasks such as FrontierSWE. On Arena.ai's blind front-end coding leaderboard, developers ranked K3 above both GPT-5.6 Sol and Fable 5, a result that will matter more to working programmers than any single benchmark number. The model is natively multimodal, handling text and images, and Moonshot is offering it through its Kimi apps and an API compatible with the OpenAI software development kit, which lowers the switching cost for anyone already building on American tools. Price is the sharper weapon. Moonshot is charging $0.30 per million cached input tokens and $15 per million output tokens, undercutting the American incumbents and echoing DeepSeek's steep discounting across the Chinese market. The release adds to an increasingly crowded open-weight field. Thinking Machines' Inkling and a run of Chinese systems have narrowed the distance to closed models from about a year to a matter of weeks. The geopolitical read is hard to miss. A Chinese lab shipping the largest freely available model, at a fraction of American prices and despite US chip export controls, is exactly the outcome Washington's restrictions were meant to forestall. Moonshot is expected to publish K3 under a modified MIT licence, the permissive terms that made Kimi and DeepSeek popular with developers wary of usage caps. The company has not confirmed the final wording, and the weights themselves remain the missing proof. The launch lands as Moonshot unwinds its VIE structure ahead of a planned Hong Kong listing. On 27 July the largest-open-model claim stops being a specification sheet and becomes something anyone can download and check.

Jensen Huang Backs 'Excellent' Chinese AI Models, Tells Washington to Ditch 'Science Fiction' Fears: 'Mis

Nvidia Corp. (NASDAQ:NVDA) CEO Jensen Huang said the U.S. should embrace, not ban, China's AI models, arguing that high-quality open-source technology should be used rather than feared. In an interview with Axios on Tuesday, Huang called the Chinese models "excellent" and dismissed concerns that OpenAI and Anthropic should fear open AI models, saying they broaden AI adoption by attracting new users, while many customers will still pay for the superior performance and reliability of closed models. Huang pushed back on fears that downloaded Chinese AI models create a "backdoor" to Beijing, arguing they can be securely customized and isolated. "There's no scenario where China runs U.S. companies off the road...Zero possibility," Huang said. He said open AI models improve security through public scrutiny and believes wider adoption of free AI will drive demand for Nvidia's chips, data centers, and computing infrastructure. "The market misunderstood the impact of DeepSeek the first time," the CEO said, adding that Wall Street has "misunderstood the impact of Kimi again this time." Huang's Advice To Washington In another part of the interview, published on Thursday, Huang urged policymakers not to let "science fiction" fears about AI shape regulation, calling predictions of mass job losses and human extinction "complete nonsense." He warned that discouraging AI adoption poses the bigger risk and encouraged officials to seek broader industry input, not just from "one or two" CEOs, before imposing restrictions. Huang said some companies may be using safety concerns to push for regulations that benefit their own interests. He also acknowledged concerns that the administration could overcorrect with excessive regulation. China AI Debate Intensifies There were also reports that the Trump Administration is planning to ban the Chinese open-source AI models after Moonshot AI released Kimi K3 last week, which outperformed several top U.S. models on certain industry benchmarks. Image via Shutterstock Market News and Data brought to you by Benzinga APIs To add Benzinga News as your preferred source on Google, click here.

NVIDIA's Jensen Quells Fears Surrounding Kimi K3 And Over-Investment In LLMs, Says "Everyone's Got It Backwards, More AI Computing Power Will Be Needed

Moonshot AI's frontier model, Kimi K3, has sent goosebumps around the industry, with one of its highlights being able to build a chip within just 48 hours that can deliver more than 8,700 tokens per second. There's also concern surrounding the AI model undercutting the likes of ChatGPT and Claude, as Microsoft has been reported to pocket $600 million in savings as it looks to swap. Fortunately, NVIDIA's Jensen Huang is here to ease all concerns, stating that the industry has the incorrect impression of Kimi K3. Jensen says people had the same idea with DeepSeek as they do with Kimi K3; a more efficient AI model won't reduce compute demand Speaking to the media shortly after attending the opening ceremony of Wistron's new plant established in Dallas, Jensen addressed Kimi K3's fears and the overall misdirecting of AI. Seeing as how Moonshot AI successfully climbed the benchmark charts with Kimi K3 while utilizing lower compute costs, the usual impression that's doing the rounds is that companies like Google, Meta, ChatGPT, Anthropic, and others had sharply increased their investments in AI to boost competitiveness, which might have been unnecessary. Jensen says that "Everyone's got it backwards, just like they did with DeepSeek. Kimi K3 is useful and very smart. More people will use it, and because of that, more AI computing power will be needed. That's the logical conclusion." Not siding with any side, Jensen mentions that the industry needs more closed models from OpenAI and Anthropic and open models from Moonshot AI, and it's not difficult to ascertain why NVIDIA's head honcho would say this, because his company stands to benefit from this boiling AI rivalry. However, he states that a model running efficiently isn't a sign of overspending from competitors, as AI adoption scales differently. NVIDIA's H200 AI GPUs, which were previously banned in China, are now reportedly heading to AI firms, opening yet another revenue stream for the graphics chip manufacturer. In short, regardless of how many efficient or inefficient models there are, the investments will continue to roll in, and when it's all over, only NVIDIA will be standing tall and walking happily to the bank. News Source: DigiTimes Follow Wccftech on Google to get more of our news coverage in your feeds.

Moonshot's Kimi K3 Needs 64 Cores for Open Source AI

The Kimi K3 model, developed by Moonshot, is poised to make waves as an open source AI system with near-frontier coding performance. Set to launch on July 27, 2026, it offers researchers and developers access to its architecture through open weights, fostering collaboration and experimentation. However, as Nate Jones explains, this openness comes with significant trade-offs. The K3 model demands substantial computational resources, 64 accelerator cores for optimal use, highlighting the scalability and accessibility challenges that often accompany open source AI initiatives. Explore how the K3 model's design reveals broader tensions between open- and closed-source AI systems. You'll gain insight into the cybersecurity risks posed by open weights, the operational inefficiencies that hinder scalability and the competitive advantages that closed-source models maintain in areas like security and innovation. This overview offers a clear lens into the evolving dynamics of AI development and the trade-offs shaping its future. What Makes the Kimi K3 Model Unique? The Kimi K3 model stands out as an open source alternative to proprietary AI systems, offering developers and researchers the ability to access and modify its architecture through open weights. This transparency fosters collaboration and innovation, making it a valuable tool for academic and independent research. However, this openness comes with significant trade-offs. The model requires substantial computational resources to operate effectively, demanding 64 accelerator cores for optimal performance. While its coding capabilities are robust, the K3 model struggles to match the efficiency, scalability and security features of its closed-source counterparts. These limitations underscore the challenges of balancing accessibility with performance in open source AI systems. Key Challenges Facing Open source AI Models Open source AI models like Kimi K3 face several obstacles that limit their ability to compete with proprietary systems. These challenges include: * High Computational Demands: The K3 model's reliance on extensive hardware resources makes it less accessible to smaller organizations and independent developers, creating a barrier to widespread adoption. * Token Inefficiency: The model consumes a large number of tokens per task, increasing operational costs and complicating efforts to scale effectively. * Performance Limitations: Despite its capabilities, the K3 model lags behind innovative closed-source systems in terms of efficiency, innovation and adaptability to complex tasks. These challenges highlight the inherent limitations of open source AI in addressing the demands of modern applications, particularly in industries requiring high levels of precision and security. Check out more relevant guides from our extensive collection on Kimi AI models that you might find useful. Why Closed-Source Models Maintain an Edge Closed-source AI models, developed by leading companies such as OpenAI and Anthropic, continue to dominate the field due to their superior performance, innovation and safety features. These systems are carefully optimized for efficiency and often incorporate advanced safeguards to prevent misuse. In contrast, open source models like Kimi K3 face difficulties in achieving similar levels of refinement and security. Chinese AI models, including Kimi K3, are estimated to trail their Western counterparts by six to seven months in development. This gap underscores the competitive advantage of closed-source systems, particularly in areas like scalability, cybersecurity and operational efficiency. The ability of closed-source models to integrate innovative technologies and maintain tighter control over their architecture gives them a distinct edge in the rapidly evolving AI landscape. Cybersecurity Risks in Open source AI The open nature of models like Kimi K3 introduces significant cybersecurity risks. By making its weights publicly accessible, the model inadvertently exposes itself to potential exploitation by malicious actors. This openness can assist activities such as hacking, fraud and other harmful applications, posing a threat to both users and broader systems. For organizations and individuals using open source AI, implementing robust cybersecurity measures is essential. Without adequate safeguards, the accessibility of open source models could lead to increased vulnerability to cyber threats. This risk underscores the importance of balancing openness with security to ensure the responsible deployment of AI technologies. Future Trends and the Role of Regulation The launch of Kimi K3 raises important questions about the future of AI development and regulation. As the cost and complexity of scaling AI models continue to rise, governments may impose stricter controls on their distribution and usage. Regulatory frameworks could play a crucial role in addressing issues such as cybersecurity, ethical deployment and equitable access. A potential solution lies in adopting a multi-model approach that combines the strengths of both open- and closed-source systems. Such an approach could enhance resilience within the AI ecosystem while mitigating risks associated with either model type. However, achieving this balance will require careful coordination between policymakers, developers and industry stakeholders. The challenge lies in fostering innovation while making sure safety, efficiency and ethical standards. Lessons for the AI Community The Kimi K3 model offers valuable lessons for developers, organizations and regulators navigating the complexities of AI development. Key takeaways include: * Emphasize AI Safety: Making sure strong safety measures is critical to prevent misuse and promote the responsible deployment of AI technologies. * Strengthen Cybersecurity: Organizations must prioritize protecting their systems from potential threats, particularly when working with open source AI models. * Prepare for Regulatory Changes: Anticipating increased government oversight and adapting strategies to align with emerging regulations will be essential for long-term success. The K3 model serves as a reminder of the importance of addressing these challenges proactively to ensure sustainable progress in the field of AI. The Broader Implications of Kimi K3 The release of the Kimi K3 model marks a significant moment in the evolution of open source AI. While it demonstrates the potential of open source approaches to foster collaboration and innovation, it also highlights their limitations in a competitive and rapidly advancing field. The model underscores the importance of balancing accessibility with efficiency, safety and innovation. As AI continues to evolve, stakeholders must navigate these challenges thoughtfully. The future of AI will depend on the ability of developers, organizations and regulators to work together in creating systems that are not only powerful but also secure, ethical and accessible. The Kimi K3 model serves as both a milestone and a cautionary tale, offering insights into the opportunities and challenges that lie ahead in the dynamic world of artificial intelligence. Media Credit: AI News & Strategy Daily | Nate B Jones Disclosure: Some of our articles include affiliate links. If you buy something through one of these links, Geeky Gadgets may earn an affiliate commission. Learn about our Disclosure Policy.

Chinese AI sensation Moonshot's gamble on big models pays off

Crowds rushed to an obscure corner of China's premier tech summit, moving past monumental booths from Alibaba Group Holding and Tencent Holdings to catch a glimpse of the hottest name in domestic artificial intelligence. Developers and users at the World AI Conference jostled for a closer look at Moonshot and its Kimi K3 -- the giant 2.8 trillion-parameter model whose release on Friday showed China was closing the gap on OpenAI and Anthropic PBC far quicker than anticipated. The startup became an instant global sensation, drawing comparisons to DeepSeek's 2025 breakout and plaudits from the likes of Tesla CEO Elon Musk. Such was the crush over the weekend that Moonshot blew through its entire stock of branded swag in just a few hours. Moonshot's emergence is a vindication not just for the country's AI industry, but also for founder Yang Zhilin. The reclusive 33-year-old was the poster child of Chinese AI in the post-ChatGPT era -- an alumnus of Carnegie Mellon University, Meta Platforms and Alphabet's Google who'd been expected to help propel the country into the big leagues. Then DeepSeek emerged, and he's toiled in its shadow ever since.

Moonshot's Kimi K3 pushes Chinese AI into Fable-level territory | Fortune

Chinese startup Moonshot AI has released the latest version of its Kimi AI model, further shrinking the performance gap between Chinese and U.S. models just as global businesses are increasingly questioning the cost of deploying models from Anthropic and OpenAI. On July 16, Moonshot AI unveiled Kimi K3, the latest version of its Kimi model. It boasts 2.7 trillion parameters, making it the largest open-weight large language model available today. (Parameters refer to the weights within the LLM; more parameters generally means that models can handle more complex reasoning.) DeepSeek V4 has 1.6 trillion parameters. "K3 stands as Moonshot AI's most powerful open-source coding model to date," Moonshot AI wrote in a press release announcing the model's release. "Operating with minimal human oversight, it can sustain long engineering sessions, navigate massive repositories, and orchestrate terminal tools." In its release, Moonshot claimed K3 performed "competitively" with Anthropic's Fable 5, currently the most advanced AI model widely available on the market today, and "substantially outperformed" Anthropic's Opus 4.8, and OpenAI's GPT 5.6 Sol and GPT 5.5. On the company's officially released benchmarks, K3 consistenty ranks within the top three models. Anthropic's Mythos 5 model, on which Fable 5 is based, is reportedly the most capable model in existence in performing cyber-related tasks, yet access to Mythos is restricted to a small number of enterprises that are part of Anthropic's Glasswing program. That program was designed to help key makers of critical infrastructure to find and patch software vulnerabilities. If K3's performance claims hold, the model would mark one of the clearest signs yet that Chinese developers can build open‑weight systems in the same class as Anthropic's and OpenAI's, with direct consequences for global competition and a fast‑evolving debate over how to regulate frontier AI. Analysts were not expecting China to produce a model as powerful as Fable until early next year. K3's release could also intensify discussions about the effectiveness of U.S. AI policy. The U.S. government temporarily imposed export controls on both Mythos and Fable after Amazon researchers found a way to jailbreak Fable's guardrails and expose Mythos' underlying cyber capabilities. It also initially told OpenAI to limit its release of GPT-5.6 to select trusted partners. The revelation that a Chinese developer created a Mythos-level model months ahead of schedule could lead to looser controls in order to ensure the U.S. companies stay ahead-or it might invigorate hawks who wish to kneecap China's AI sector as much as possible. U.S. politicians are considering ways to stop Chinese developers from "distilling" U.S. AI models, which is when the outputs of a larger, more powerful AI model is used to help train smaller, more efficient models. Anthropic has accused Moonshot, z.ai, Minimax, Alibaba and DeepSeek of "illicit" distillation attacks. U.S. officials are also discussing ways to curb the appeal of open-source models from China, perhaps by encouraging the creation of U.S. open-source models. Chinese AI models are winning converts around the world, due to their lower cost and greater efficiency. Also, as open-source models, developers can download the models for free and tweak them to suit their own purposes. But using open source models does often require more technical expertise on the part of the companies deploying them. It also requires those companies to rent AI chips through cloud providers in order to host the models. U.S. export controls barred Chinese developers from getting access to the advanced AI processors used to train and run the most powerful AI models. That forced Chinese developers to find new ways to get more bang for their computing buck. "We knew we didn't have the luxury to simply scale up compute," Yutong Zhang, president of Moonshot AI, said at the World Economic Forum earlier this year. "That forced us to focus on fundamental research and efficiency." Moonshot's previous AI models were already making inroads into Silicon Valley. Cursor, the vibe-coding startup, used Kimi to help build Composer 2, its AI coding agent; Doordash also delegates "lower-level work to Kimi K2.6," according to chief technology officer Andy Fang in an early July social media post. Thinking Machines also tapped Kimi K2.5 to generate early post-training data for its new Inkling model, released on July 15. K3 is expensive -- by Chinese standards. K3 costs $15 per million output tokens, compared to $4.40 per million output tokens for z.ai's GLM-5.2 and $0.87 for DeepSeek V4. Still, it's cheaper than the equivalent U.S. models: Fable costs a whopping $50 for the same amount of output. Moonshot AI raised $2 billion in funding in May, valuing the company at over $20 billion. A statement from the company's financial advisor stated Moonshot's annual recurring revenue exceeded $200 million. Moonshot's backers include all of China's largest tech firms -- including Alibaba, Tencent and Meituan -- as well as Hongshan Capital. Moonshot's fellow AI developers, MiniMax and z.ai, went public in Hong Kong in early January. Moonshot AI, too, is reportedly preparing for an initial public offering in Hong Kong; DeepSeek, in contrast, is considering a listing in Shanghai.

Chinese AI model takes US tech industry by surprise with abilities rivalling Claude, ChatGPT

The newest Kimi K3 model from Beijing-based startup Moonshot, run by a Pink Floyd-loving entrepreneur who earned his doctorate in Pittsburgh, appears to be catching up to the best versions of Anthropic's Claude and OpenAI's ChatGPT. Another powerful new artificial intelligence model from China took the US tech industry by surprise Friday, the latest sign that Chinese startups that publicly release their "open-source" AI technology are making the California titans of AI sweat. The newest Kimi K3 model from Beijing-based startup Moonshot, run by a Pink Floyd-loving entrepreneur who earned his doctorate in Pittsburgh, appears to be catching up to the best versions of Anthropic's Claude and OpenAI's ChatGPT. "This may be the single biggest release of the year," and marks a moment when open-source Chinese models are surpassing closed US models, said Anastasios Angelopoulos, co-founder and CEO of Arena, a platform for evaluating AI systems. Kimi K3 topped the charts on Arena's ranking of what it calls "front-end coding capability," which is one measure of an AI large language model's performance. "More results are rolling in that are likely to continue to show it is at the top of the pack," Angelopoulos said on social media. It was not likely a coincidence that K3's unveiling came shortly before Chinese President Xi Jinping's opening address Friday to the nation's annual World Artificial Intelligence Conference in Shanghai. American-led restrictions have blocked China from accessing some of the world's most advanced technologies, spurring China's efforts to build its own know-how and intensifying the rivalry between the world's two biggest economies. "The development of artificial intelligence should not be a solo performance by any single country but rather a symphony of global cooperation," Xi said at the event. Chinese AI models have shown large strides K3 follows another major AI model release last month from the Chinese startup Zhipu, or Z.ai. Its new flagship GLM-5.2 model is already being widely used by software developers around the world who say it can perform work almost as good as the top US models at a cheaper price. The hype over the new Chinese model resembles the market-shaking panic that followed the release of a new model from Chinese startup DeepSeek in early 2025, though not everyone finds it justified. The response to K3 is an "overreaction shockingly similar" to DeepSeek's release last year, said tech analyst Patrick Moorhead on social media. He said it could be good for parts of the broader AI industry but pose a revenue challenge to Anthropic and OpenAI. During the conference that runs until Monday, tech giant Huawei has also been showcasing a new AI computing system called the Atlas 950 SuperPoD, a signal that China increasingly is amassing the domestic hardware it needs despite US restrictions on imports from chipmakers like Nvidia. Moonshot hasn't said what hardware it used to build K3, but the startup is a partner with Huawei. The price to use K3 is highest yet for a Chinese AI model, but it is still half as expensive as OpenAI's high-performing GPT-5.6 Sol model, according to a Friday report by Bank of America research analysts. US politicians and several major US AI companies including Anthropic and OpenAI have accused Chinese AI models of illicit "distillation" of their models to extract their technologies, a claim that Beijing says is "groundless." Anthropic in February accused DeepSeek, Moonshot and a third China-based AI lab, MiniMax, of engaging in campaigns to "illicitly extract Claude's capabilities to improve their own models" using the distillation technique that "involves training a less capable model on the outputs of a stronger one." Anthropic said that distillation can be a legitimate way to train AI systems but it's a problem when competitors "use it to acquire powerful capabilities from other labs in a fraction of the time, and at a fraction of the cost, that it would take to develop them independently." But it can go both ways. San Francisco-based startup Anysphere, maker of the popular coding tool Cursor, has acknowledged that one of its top products was based on Moonshot's K2.5 model. Elon Musk's SpaceX is planning to close a deal to buy Cursor for USD 60 billion later this year. K3 marks a 'leap for open-source' AI models Moonshot co-founder and CEO Yang Zhilin earned his Ph.D. in 2019 at Carnegie Mellon University, where he is said to have made fundamental contributions to the machine-learning field and was known for a love of rock bands like Pink Floyd. The pride among his former colleagues at the school in Pennsylvania transcends the US-China rivalry. "What a huge win for the open-source community! It feels like just yesterday Zhilin was graduating from my lab at CMU," wrote his former adviser Russ Salakhutdinov, who is also a former director of AI research at Apple. Developers who build "open-source" AI make key components of the technology accessible for anyone to examine, modify and build upon. Proponents say open-source practices promote innovation, while critics warn that making powerful AI models publicly accessible poses safety and security dangers.

China's Moonshot Challenges Anthropic With a Bigger, Cheaper Model | PYMNTS.com

Moonshot is expected to release Kimi K3 in the coming days, the Financial Times reported, citing two people familiar with the matter. The model is expected to have between 2 trillion and 3 trillion parameters, making it China's largest AI model to date. Industry analysts estimate Anthropic's Claude Opus 4.8 has between 1.5 trillion and 2 trillion parameters, though Anthropic has not disclosed the figure. According to the report, Kimi K3 is expected to outperform Claude Opus 4.8 on mainstream benchmarks and will be released as an open-weight model, allowing developers to download and modify it. The model is still expected to trail Claude Fable 5, Anthropic's most powerful model, which was briefly withdrawn after U.S. officials raised concerns about its cybersecurity capabilities. The launch underscores how quickly Chinese AI developers are narrowing the technology gap with U.S. rivals. The Financial Times reported that Kimi K3 could challenge the industry's long-held assumption that Chinese frontier models trail their American counterparts by eight to 12 months. The competitive pressure increasingly extends beyond benchmark performance to economics. PYMNTS reported that Chinese developers, including DeepSeek, have gained enterprise interest by offering AI models at a fraction of the cost charged by leading U.S. providers, forcing businesses to reconsider whether premium frontier models justify their higher prices. According to the Financial Times, Anthropic plans to raise prices for Claude Opus 4.8 in September to $3 per million input tokens and $15 per million output tokens. By comparison, Moonshot's current K2.6 model costs roughly one-third as much, while remaining open-weight rather than proprietary. The rivalry has also intensified concerns over intellectual property. The Financial Times noted that Anthropic accused Chinese AI companies earlier this year of conducting "industrial-scale distillation attacks," in which developers train smaller models using outputs from frontier systems instead of building them entirely from scratch. The competitive race is also reshaping AI investment. PYMNTS reported that OpenAI is weighing delaying its IPO until 2027 because advisers are concerned that technology stock volatility could weaken investor demand. Meanwhile, Chinese AI companies continue raising capital, with Moonshot reportedly seeking a valuation of about $31.5 billion while rival DeepSeek is pursuing a valuation of roughly $71 billion, according to the Financial Times.

Bessent Wants Chinese AI Sanctioned, But Jensen Huang Says Distillation Is Learning - NVIDIA (NASDAQ:NVDA

The pressure started Tuesday, when Treasury Secretary Scott Bessent told Fox Business the administration is finding "watermarks of our U.S. large language models on many of the Chinese models" and may act within weeks. The practice at issue is distillation: bombarding a frontier model with prompts and training a cheaper model on its answers, capturing much of its capability at a fraction of the cost. The technique is a normal part of AI development when done openly. The White House alleges Moonshot did it covertly, at industrial scale, in violation of Anthropic's terms. Huang sees it differently. "Distillation, learning from AI, learning from other sources of knowledge, is fundamental to intelligence," he told Axios. He added that U.S. labs have little to fear from open rivals. "There's no scenario where China runs U.S. companies off the road," Huang said. "Zero possibility." The White House Names Names Hours after Huang's comments published, White House science chief Michael Kratsios escalated. "We have information that Moonshot AI distilled Anthropic's Fable for the development of its K3 model," he posted on X, alleging Moonshot built an internal platform that switched access methods to avoid detection and accessed Nvidia GB300 chips in Thailand. Anthropic reportedly traced 3.4 million exchanges with its models to Moonshot earlier this year, routed through hundreds of fake accounts. Anthropic itself recently had a $1.5 billion settlement approved over pirated books used in training. The accusation lands days after Kimi K3's release rattled chip stocks and prompted David Sacks and Bill Ackman to sound the alarm on America's narrowing AI lead. Huang's logic may be commercial. Cheap open models expand AI's audience, he argued, which likely drives more demand for the chips and data centers Nvidia sells. What Prediction Markets Say Anthropic, whose Fable 5 model just disproved a math conjecture that stood for 87 years, trades at 65% on Polymarket to hold the best AI model at year-end, with Moonshot the top Chinese lab at 2%. The next flashpoint may come in September, when Bessent is expected to lead the first official U.S.-China AI dialogue ahead of Xi Jinping's planned visit. NVDA Stock Price Activity: NVIDIA shares were up 3.10% at $213.72 at the time of publication on Wednesday, according to Benzinga Pro data. Image: Shutterstock Market News and Data brought to you by Benzinga APIs To add Benzinga News as your preferred source on Google, click here.

An OpenAI Exec Thinks Kimi K3 And Other Open-Weight Models Are Bringing On "AI Communism"

OpenAI is predictably feeling the heat from Kimi K3's shock elevation to the top of Arena's front-end code rankings, and is now resorting to xenophobia and 'intellectual elitism' to try to protect its CapEx-heavy turf. OpenAI's head of strategic futures responds to Kimi K3's ascendancy by theorizing that an open-weight-model-dominant world is equivalent to "full AI communism" As we explained in a recent post, Moonshot has just unveiled its brand-new, open-source Kimi K3 model, which spans 2.8 trillion parameters, and is designed specifically for frontier-scale intelligence. The model is multi-modal, has a context window of around 1 million tokens, and offers competitive and faster performance than many of its compeers. While Moonshot likely distilled the model's foundations from Anthropic's Claude, the Kimi K3 does sport a novel KV cache-related architecture, called Kimi Delta Attention or KDA. Consider a scenario: you are writing a story, but hampered by terrible short-term memory. Whenever you write a new word, you are compelled to read whatever you've written so far just to remember what has already been inked. Obviously, as the text length increases, so does this laborious process. Key-Value or KV cache is similar to taking notes on a separate sheet so that you remain abreast of what has been written so far. This speeds up the entire process by orders of magnitude. Unlike the prevailing quadratic-attention approach, where the KV cache scales in a linear manner with the given context, Kimi K3 adopts a hybrid linear-attention mechanism, which uses a Recurrent and a Hybrid Interleaving state in a 3:1 ratio. To simplify this concept, remember that the conventional approach requires the model to write down every sentence that it reads on a separate stack of sticky notes, which means that the KV cache becomes larger as context increases, slowing down the overarching processing. Kimi K3's approach, however, involves using three assistants (Recurrent states), where each assistant keeps a running summary of the really important stuff on a single sticky note page. No matter how long the context becomes, the summary size does not increase as the older, more stale information is erased. This, however, will inevitably lead to these assistants missing out on a very specific tiny detail (like a phone number on page 100). To rectify this shortcoming, the fourth assistant (called the Hybrid Interleaving state) takes a photo snapshot of the entire context and then compresses it, akin to compressing a 4K movie into a tiny zip file. Whenever the model feels that it needs to look at that 'global snapshot,' it inflates the file momentarily, takes a peek, and then compresses it again. This is the pattern that then emerges: This pattern keeps repeating over and over again. And because 3 out of the 4 assistants use an ultra-fast memory-layer approach, the KV cache is dramatically reduced. While the KDA makes Kimi K3 very efficient, the model's overall use of the HBM is not expected to decline due to the use of another optimization technique, called WideEP. These Kimi K3 innovations, however, are a direct threat to OpenAI and Anthropic. And now, OpenAI's head of strategic futures, Dean W. Ball, has penned an X post, calling all Chinese open-source AI models "decelerationists" for CapEx, going on to declare that a "probable outcome of an open-weight-model-dominant world is full AI communism, which is precisely what China proposes: rather than a market product, AI is a "public good" which will ultimately be provided by the state." Meanwhile, Anthropic has responded by retaining access to its top-of-the-line Fable 5 model across major subscription tiers now that Kimi K3 has overshadowed Opus. Follow Wccftech on Google to get more of our news coverage in your feeds.

Kimi K3 Open Source AI Rivals GPT 5.6 Sol with 2.8T Parameters

The Kimi K3, developed by Moonshot AI, stands out in the AI landscape with its impressive 2.8 trillion parameters and a 1 million token context window. These specifications enable it to tackle complex tasks requiring both depth and scale, such as generating detailed narratives or designing intricate 3D environments. According to World of AI, the Kimi K3 also supports multimodal processing, seamlessly handling text, images and other data types, making it a versatile option for applications like coding, game development and creative workflows. Explore the features that distinguish the Kimi K3, including its affordability and how it enables broader experimentation with high-performance AI. Learn about its ecosystem of integrations, such as the Kimiko API and K3 Swarm, which enhance its functionality across technical and creative domains. Gain insight into how this model compares to proprietary systems like GPT 5.6 Sol and Claude Fable 5, offering a comprehensive view of its capabilities and potential applications. Features That Set the Kimi K3 Apart The Kimi K3's architecture is carefully designed to handle complex and nuanced tasks with remarkable precision. Its standout features include: * 2.8 Trillion Parameters: This vast parameter count enables the model to deeply understand and generate intricate patterns in data, enhancing its ability to tackle sophisticated tasks. * 1 Million Token Context Window: The extended memory capacity allows the Kimi K3 to process and analyze lengthy documents, generate detailed narratives and manage tasks requiring long-term contextual understanding. * Multimodal Integration: The model seamlessly processes text, images and other data types, making it versatile for applications such as coding, 3D modeling and creative design. These features empower the Kimi K3 to address challenges traditionally dominated by proprietary systems, leveling the playing field for open source AI and expanding its potential applications. Performance That Rivals Proprietary Models The Kimi K3 has demonstrated its ability to compete with industry-leading proprietary models, ranking third on the World of AI benchmark. Its performance is particularly notable in several key domains: * Coding: The Kimi K3 outperforms Claude Opus 4.8 in coding benchmarks, showcasing its ability to generate efficient, reliable and optimized code for various programming tasks. * Game Development: It competes closely with GPT 5.6 Sol and Fable 5 in creating complex games, including those with advanced physics and interactive mechanics. * 3D Design: The model excels in producing polished 3D environments and interactive systems, delivering remarkable detail and accuracy for design projects. This level of performance highlights the Kimi K3's versatility and technical strength, making it a valuable tool for developers, designers and researchers seeking high-quality results across a range of disciplines. Unlock more potential in open source AI by reading previous articles we have written. Empowering Creativity and Technical Precision The Kimi K3 is uniquely designed to excel in tasks that demand both creativity and technical expertise. Its advanced capabilities include: * Generating immersive 3D environments for simulations, virtual reality projects and gaming applications. * Developing complex games with realistic physics, interactive mechanics and engaging narratives. * Simulating virtual hardware environments, such as macOS clones, for testing and development purposes. * Producing high-quality SVG outputs and interactive web designs tailored for creative and technical projects. These features make the Kimi K3 a comprehensive solution for users seeking to push the boundaries of innovation, whether in creative industries, technical development, or research. Cost Efficiency Without Sacrificing Quality One of the most compelling aspects of the Kimi K3 is its affordability. Priced at $3 per 1 million input tokens and $15 per 1 million output tokens, it offers significant cost savings compared to proprietary models. This pricing structure democratizes access to high-performance AI, allowing individuals, startups and organizations to experiment, innovate and create without the financial barriers often associated with advanced AI systems. By providing innovative capabilities at a fraction of the cost, the Kimi K3 ensures that high-quality AI tools are accessible to a broader audience, fostering innovation across diverse sectors. A Comprehensive Ecosystem for Diverse Applications The Kimi K3 is supported by a robust ecosystem of tools and integrations, enhancing its usability and adaptability across various domains. Key components of this ecosystem include: * Kimi Chatbot: A conversational AI tool designed for browser-based tasks, customer interactions and real-time communication. * K3 Swarm: A collaborative framework that facilitates distributed computing and problem-solving, ideal for large-scale projects. * Kimiko API: A versatile interface that allows seamless integration of the Kimi K3 into coding, design and creative workflows. These tools expand the Kimi K3's functionality, making it suitable for a wide range of use cases, from developing interactive systems to simulating virtual environments. This ecosystem ensures that users can fully use the model's capabilities to meet their specific needs. Shaping the Future of Open source AI The Kimi K3 is more than just a high-performing AI model; it represents a paradigm shift in the open source AI ecosystem. By delivering state-of-the-art performance at a fraction of the cost of proprietary systems, it challenges the notion that advanced AI must remain exclusive to closed platforms. The Kimi K3 sets a new standard for what open source AI can achieve, fostering innovation, collaboration and accessibility across the global AI community. This model not only redefines the potential of open source AI but also enables users to explore new frontiers in artificial intelligence, bridging the gap between affordability and innovative technology. Whether you are a developer, designer, or researcher, the Kimi K3 offers a powerful platform for tackling complex challenges and unlocking new possibilities in AI-driven innovation. Media Credit: WorldofAI Disclosure: Some of our articles include affiliate links. If you buy something through one of these links, Geeky Gadgets may earn an affiliate commission. Learn about our Disclosure Policy.

Moonshot AI's Kimi upends conventional wisdom on U.S. lead over China

At an event in Beijing earlier this year, some of China's top artificial intelligence leaders warned that the country remained meaningfully behind the U.S. in developing cutting-edge AI models, with one executive arguing that "the gap may actually be widening." Leading U.S. firms also appeared confident they were significantly ahead. As recently as last week, one executive at Anthropic, who spoke on condition of anonymity, mused that the Claude maker's technology was roughly six to 12 months ahead of Chinese rivals. On Friday, Moonshot AI upended those assumptions. The Chinese AI lab released Kimi K3, a more advanced open-weight model that it said outperforms all rivals except for Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 on overall capability. The implication is that Moonshot, and by extension China, could be closing the gap faster than expected.

David Sacks warns U.S. could lose AI race after Chinese model's breakout

Why it matters: Sacks is pitching a hands off regulatory approach to AI that he frames as do-or-die for America to hold onto its dominance. What they're saying: "This is concerning," Sacks wrote on X, noting that for the first time a Chinese model has taken the top spot on the Frontend Code Arena, an AI evaluation platform, and is "scoring at or near the frontier" on other benchmarks. * Meanwhile, he argued, America is "tying itself in knots" with its policy reactions to AI, including bans on new data centers and the recent push for federal agencies to pre-approve model releases. * "This is how you lose the AI race," he wrote. "The rest of the world won't play by our rules if we bog ourselves down." * Sacks dropped the official title of Trump's AI Czar, but retains the influence. He's currently the co-chair of the President's Council of Advisors on Science and Technology. Catch up quick: Kimi K3, released Thursday by Beijing-based Moonshot AI, beat Anthropic's Fable 5 and OpenAI's GPT-5.6 Sol in blind front-end coding tests and outranked Opus 4.8 in Arena's broader text ranking, at roughly 40% lower cost. * Moonshot plans to release the model's weights July 27, letting companies and governments customize and run it on their own systems. Open-weight models are having a moment. * Cost concerns, regulatory uncertainty and enterprise demand for more control over models are all driving greater business interest to open-weight models. Zoom out: Kimi's release lands squarely in the middle of a live Washington fight, as calls grow for regulation of frontier models and CEOs argue they can't slow down for fear that China will catch up. * Sacks's prescription is the same one he's championed for years. * "Permissionless innovation is how America won the internet," he wrote, arguing the U.S. can win AI the same way "while addressing risks in a targeted way." Friction point: Sacks' call for "targeted" safeguards contrasts with Anthropic's push for a statutory process that would allow the government to block unsafe frontier-model deployments. * Just five weeks ago, Anthropic proposed giving the government standing authority to block frontier model releases that fail independent safety tests, with penalties tied to global revenue. * Sacks's post can be read as a direct response to that: Kimi K3, he's arguing, is what happens while America builds approval gates that Beijing will never walk through. Yes, but: U.S. regulators cannot directly control a Chinese lab's release of an open-weight model. * Once the weights are publicly available, companies and governments may be able to download, customize and run the model themselves. The bottom line: Sacks is using Kimi K3 to argue that U.S. AI regulation could give Chinese developers a competitive opening.

Race for AI dominance heats up as China releases high-performance open-source models

The US-China technology rivalry, once focused largely on advanced chips and semiconductor manufacturing equipment, is now expanding into a broader contest over artificial intelligence models and the future rules of global AI governance. As Chinese developers close the gap with leading US AI systems, the Trump administration is reviewing measures to restrict US companies from using Chinese AI models, while Beijing is seeking to challenge the US-centered AI ecosystem through open-source development and international cooperation. Axios reported Monday (local time) that Washington has begun to review measures to bar US companies from utilizing Chinese AI models following Friday's launch of Kimi K3, Chinese startup Moonshot AI's open-source AI model. Kimi K3 has sent shockwaves throughout the AI sector for performing on par with flagship models released by market dominators Anthropic and OpenAI. Some of the measures the US administration is considering are said to involve export controls to restrict the use of Chinese AI models or publicly adding pressure to US companies utilizing Chinese models. Until now, the US has attempted to curb China's AI development by sharply restricting semiconductors and chip manufacturing equipment exports to China, but it has not outrightly banned commercial use of Chinese AI models. But Kimi K3 isn't the only Chinese AI model that is causing an uproar. Just two days after Moonshot AI released its latest model, competitor Alibaba opened up a preview version of its Qwen3.8 Max, its flagship model, to developers on Sunday. Alibaba called the model, which has 2.4 trillion parameters, comparable to frontier models and second only to Fable 5 from Anthropic. Considering that it hasn't even been two full months since Fable 5's release in early June, industry observers suggest that China may have narrowed the tech gap between it and the US from 6-9 months to 2-3 months. Chinese AI models are also gaining ground by leveraging lower token costs. According to monthly usage rankings on OpenRouter, an AI model aggregation platform, the top six models as of Tuesday were all developed by Chinese companies, including Xiaomi, DeepSeek, Tencent, MiniMax, and Zhipu AI. Most of those models are open-source, meaning that they make their underlying parameters publicly available, allowing them to offer lower token costs than leading closed-source models from US companies. OpenRouter's State of AI 2025 report, released late last year, showed that Chinese models accounted for just 1.2% of weekly usage share at the end of 2024. That figure rose to an average of 13% in 2025, reaching nearly 30% some weeks. China is using the rapid advancements in its homegrown models as a basis to stake its place in the competition for AI dominance and norm-setting. During a keynote address at the opening ceremony of the World AI Conference last week, Chinese President Xi Jinping said that AI development "should not be a solo performance by a single country, but a symphony of international cooperation," demonstrating Beijing's advocacy for openness and cooperation -- a far cry from the US move to consider restrictions on exports of its models to China as well as curbs on use of Chinese models within the US. When China established the World Artificial Intelligence Cooperation Organisation at the conference last week, 29 countries signed on as founding members, including Brazil and Russia. "The remarks amounted to Xi's clearest articulation yet of China's ambition to shape global AI governance, framing its open-source models as a global public good and positioning Beijing as an alternative to Washington at a pivotal moment in the race for technological leadership," Reuters said of the Chinese leader's speech. By Kang Jae-gu, staff reporter

Moonshot AI's Kimi K3 Pushes China Into Anthropic's Fable 5 Territory

The latest launch could cause ripples among US enterprises who are seeking cheaper options for their AI interface Chinese startup Moonshot AI has released the latest version of its Kimi foundational model which it claims would further shrink the performance gap between US and Chinese models. The launch comes at a time when global enterprises, including the likes of Microsoft, are questioning the cost of deploying models from Anthropic and OpenAI and seeking alternatives. The Chinese company announced last night that it had unveiled Kimi K3 that boasts of 2.7 trillion parameters, making it the largest open-weight LLM today. As readers would be aware, this measure refers to weights within the model and more the parameters, the more that model can handle complex reasoning. Just for comparison, DeepSeek V4 has 1.6 trillion parameters. In a statement released on its website, Moonshot AI says, ""K3 stands as Moonshot AI's most powerful open-source coding model to date. Operating with minimal human oversight, it can sustain long engineering sessions, navigate massive repositories, and orchestrate terminal tools." The release claimed that K3 had performed "competitively" with Anthropic's Fable 5, yes the same model that President Trump had blocked from the export market. It also claimed that the model had "substantially outperformed" Anthropic's Opus 4.8, OpenAI's GPT-5.6 Sol and GPT-5.5 among others. On its own benchmarks, K3 consistently ranked among the top-3. "While its overall performance still trails the most powerful proprietary models, Claude Fable 5 and GPT 5.6 Sol, Kimi K3 demonstrated frontier-level performance across our evaluation suite, consistently outperforming other tested models," the company disclosed. Of course, K3 could steal a march over Anthropic's Mythos 5 model on which the heavily guard-railed Fable5 is based. Given that Anthropic has itself restricted Mythos to a small number of enterprises as part of its Glasswing program, Moonshot may well make a mark with companies that are questioning the hegemonistic approach of the Claude-maker and OpenAI. Moonshot AI says "Kimi K3 is the first open model to reach 2.8 trillion parameters. It marks the latest step in Kimi's sustained push at the scaling frontier: for nine of the past twelve months, Kimi models have set the upper bound of open-model sizes." The company has made it available today via Kimi.com, Kimi Work, Kimi Code, and the Kimi API. Moonshot says at launch Kimi K3 will use max thinking efforts by default with low and high-effort modes getting introduced in subsequent updates. "We are currently working closely with inference partners and open-source maintainers to align technical details and ensure a reliable rollout across the ecosystem," the company said. The full model weights will be released by July 27, 2026. Further details on the architecture, training, and evaluations will be released alongside the Kimi K3 technical report. Now, if Kim K3's performance holds good, the model could be the first to indicate that China can build open-weight systems in the same class as Anthropic and OpenAI, thus bridging the US-China gap. We had reported yesterday on how Mira Murati's Thinking Machines Lab is already using Kimi K2.5. The company had accepted some levels of data distillations for its Inkling open weight model. They said it was pre-trained from scratch and used open-weight models like Moonshot's AI's Kimi K2.5 to generate some early post-training data.

Kimi K3 Built A Chip In Just 48 Hours, Which Pushes Over 8700 Tokens/s, As China's Moonshot Delivers A 2.8 Trillion Parameter Frontier AI Model

Moonshot has unveiled its latest Kimi K3 Open model, which packs a stunning 2.8 trillion parameters & was able to build a chip that crunches 8700 tokens/s in inference. China's Kimi K3 AI Model Autonomously Designed A Fully Functional Chip That Delivers Over 8700 Tokens/s Through Inference Workloads Today, China's leading AI firm, Moonshot, unveiled its brand-new Kimi K3 model, which is the world's first 3T-class "Open" AI model, spanning 2.8 trillion parameters, designed specifically for frontier-scale intelligence. The technical highlights of Kimi K3 include: * Parameters: 2.8 trillion total (Mixture-of-Experts / MoE architecture) * Context window: 1,048,576 tokens (1M) * Multimodal: Native support for text + image + video * Architecture highlights: Innovations including Kimi Delta Attention (KDA) + Attention Residuals (claimed up to 6.3× faster decoding at million-token contexts) * Weights format: MXFP4 weights / MXFP8 activations (quantization-aware from training) * Variants at launch: K3 Max (general/chat/agent) and K3 Swarm Max (large-scale parallel processing) * Reasoning: Always-on (with configurable effort via API at launch) * Open weights: Promised release by ~July 27, 2026 (first open ~3T-class model) * Pricing (API): $3 / $15 per million tokens (input/output); cached input is significantly cheaper (~$0.30) * Availability: Kimi app, Kimi Code, Moonshot OpenAI-compatible API, OpenRouter * Benchmarks (self-reported by Moonshot): Strong in agentic/coding tasks (e.g., Terminal-Bench 2.1: 88.3, just behind GPT-5.6 Sol). Competitive overall but trails top closed models in broad evaluations. The AI model is said to trail behind flagship solutions such as OpenAI's GPT 5.6 and Anthropic's Claude Fable 5, but the model is available today and delivers some interesting highlights, such as a chip design that it was able to complete autonomously within 48 hours. The main highlights of the AI model include: * Autonomous chip design -- The model designed a real, functional chip (using its own architecture) in a single 48-hour autonomous run with zero human intervention in the design. It handled design, optimization, verification, timing, and simulation. The resulting chip (4 mm² area, Nangate 45nm library, open-source EDA tools) achieved > 8,700 tokens/second decoding throughput in simulation. * Built a GPU compiler from scratch -- created MiniTriton (a custom GPU programming system/compiler) entirely from scratch. It reportedly beats or matches parts of NVIDIA's official Triton compiler on benchmarks. * Professionally edited its own launch video -- It took 56 raw video clips and edited a polished teaser, handling clip selection, action continuity, frame-level beat synchronization to music, audio processing, and multiple revision rounds. It also reportedly created 3Blue1Brown-style motion graphics explaining its own architecture. In the various public benchmarks, Kimi K3 offers competitive and faster performance than the competition thanks to broad improvements in Kimi K3's agentic knowledge work capabilities, enabling more capable and reliable performance in real-world use cases. 8700+ Tokens/s Chip Made In Just 48 Hours One of the highlights of Moonshot's Kimi K3 is that it was able to build a chip using its own model for its own Nano model. The chip is based on Kimi's own architecture, and was completed within a 48-hour autonomous run, which included the tape-out of the silicon "K3 built", optimization, and verification using open-source EDA tools (Nangate 45nm library). The chip itself measures 4mm2 and clocks in at around 100 MHz. It packs 1.46 million standard cells, 0.277MB of SRAM, and an INT4 MAC array which features fused dequantization. In initial tests, the chip was able to achieve over 8700 Tokens/s of AI inference performance (8721 Tokens/s to be precise). The Future of Game Dev & Digital Content Creation Kimi K3 also promised to deliver strong and exceptional results for 3D reasoning, coding, & vision capabilities. It can be used to transform images and videos into fully playable interactive experiences, paving the way for future games where entire experiences can be built upon using a reference image or video. The reference demo shows a character riding a horse through a small farm with trees in the back of the environment and a small lake. Rain is present in the scene, and the model does a great job in generating the scene, though this is just an early example. Moonshine shares several examples where images and videos can be turned into playable and interactable experiences. Some of the examples of generated games/experiences can be seen below: Kimi K3 represents a major leap in open-source AI, delivering frontier-level intelligence in long-horizon coding, multimodal reasoning, and complex knowledge work through its innovative 2.8T-parameter architecture. As the world's first open 3T-class model, it narrows the gap with leading proprietary systems while excelling in agentic tasks -- from GPU kernel optimization and full compiler development to interactive research visualizations and creative video editing. With native vision, a 1-million-token context, and broad availability today (full weights releasing July 27, 2026), Kimi K3 empowers developers, researchers, and creators with unprecedented open access to advanced AI capabilities. Follow Wccftech on Google to get more of our news coverage in your feeds.

Leaked Kimi K3 Reportedly Beats GPT 5.6 Sol in Early Tests

The AI landscape is buzzing with the unexpected leak of Kimi K3, a model developed by Moonshot AI under the codename "Keyine." According to Universe of AI, early benchmarks reveal that Kimi K3 excels in areas like spatial reasoning and 3D generation, outperforming even top-tier systems such as GPT 5.6 Sol and Fable 5. For instance, its ability to interpret intricate environments positions it as a standout candidate for robotics and simulation applications. This combination of technical precision and versatility underscores its potential to influence a broad range of industries. Dive into this exposé to explore how Kimi K3's standout features could reshape AI applications. Gain insight into its agentic capabilities, which enhance autonomous decision-making in collaborative tasks and its affordability, which makes advanced AI accessible to a wider audience. Additionally, learn about the growing speculation around its potential release as an open-weight model, a move that could foster innovation and collaboration across the AI community. These developments highlight why Kimi K3 is being closely watched as a pivotal moment in AI evolution. Outperforming the Competition Initial evaluations highlight Kimi K3's ability to excel in areas where even the most advanced AI models often encounter limitations. Its standout performance in spatial reasoning, agentic capabilities and 3D generation has positioned it as a formidable competitor. For example, in spatial reasoning tasks, Kimi K3 demonstrates an exceptional ability to interpret and navigate intricate environments. This makes it particularly well-suited for applications in robotics, simulation and interactive AI systems. When compared to established models like GPT 5.6 Sol and Fable 5, Kimi K3's results suggest it could redefine industry benchmarks. Its capacity to handle complex tasks with precision and efficiency underscores its potential to lead in both technical and creative domains. By addressing challenges that have traditionally limited AI performance, Kimi K3 sets a new standard for what advanced AI systems can achieve. Affordable AI Without Compromise One of the most compelling aspects of Kimi K3 is its affordability. Moonshot AI has designed the model to be accessible to a broad spectrum of users, ranging from researchers to small businesses, without compromising on performance. This approach could provide widespread access to access to advanced AI, empowering more individuals and organizations to harness innovative technology without the financial strain typically associated with premium models. Speculation is growing that Kimi K3 may be released as an open-weight model. If confirmed, this would allow developers and researchers to access its architecture and weights, allowing them to customize and optimize the model for specific use cases. Such a move could significantly expand its reach, fostering innovation across diverse industries and encouraging collaborative advancements in AI development. Explore further guides and articles from our vast library that you may find relevant to your interests in Kimi AI models. Key Innovations in Kimi K3 Kimi K3 introduces several new advancements that set it apart from its competitors. These innovations include: * Spatial Reasoning: An unparalleled ability to interpret and navigate three-dimensional spaces, making it ideal for applications in gaming, virtual reality and autonomous systems. * Agentic Capabilities: Enhanced autonomy in decision-making, particularly in scenarios requiring collaboration among multiple AI agents to achieve complex objectives. * 3D Generation: Superior creation of three-dimensional models and environments, allowing immersive and interactive experiences across industries. * Pixel Generation: High-quality visual outputs that rival human artistry, offering immense value to creative industries such as media production and design. These features not only enhance the model's functionality but also broaden its potential applications. From industrial automation to entertainment, Kimi K3's capabilities open doors to new possibilities in AI-driven innovation. Shaping the AI Market The emergence of Kimi K3 is poised to intensify competition within the AI industry. By offering a high-performance, cost-effective alternative, Moonshot AI is directly challenging established players like OpenAI and Anthropic. This development echoes the fantastic impact of past breakthroughs, such as the "Deepseek R1 moment," which reshaped the AI landscape and set new standards for innovation. For users, this heightened competition could lead to accelerated advancements and potentially lower costs. If Kimi K3 is released as an open-weight model, it could further drive progress by allowing a global community of developers to build upon its foundation. This collaborative potential may usher in a new era of AI research and development, fostering a more inclusive and dynamic ecosystem. What Kimi K3 Means for You Kimi K3's unique combination of high performance, affordability and open-weight potential positions it as a significant development in the AI industry. Whether you are a developer seeking a customizable model, a business in need of cost-effective AI solutions, or a researcher exploring new frontiers, Kimi K3 offers a compelling option tailored to diverse needs. As Moonshot AI moves closer to the official release of Kimi K3, the industry is watching with anticipation. This model has the potential to redefine expectations for what AI can achieve, making it a pivotal development in the rapidly evolving world of artificial intelligence. With its innovative features and accessible design, Kimi K3 is undoubtedly a model to watch as it shapes the future of AI technology. Media Credit: Universe of AI Disclosure: Some of our articles include affiliate links. If you buy something through one of these links, Geeky Gadgets may earn an affiliate commission. Learn about our Disclosure Policy.

China's Moonshot unveils world's largest open AI model, closing in on U.S. rivals

Chinese AI startup Moonshot on Friday unveiled Kimi K3, a 2.8 trillion-parameter model that it said is the world's largest open-weight AI system and delivers performance approaching U.S. giant Anthropic's frontier Fable model. The launch, which comes a month after Anthropic's Fable and Mythos models were abruptly withdrawn by the U.S. government due to security concerns, underscores how quickly China's open AI ecosystem is narrowing the gap with the most advanced U.S. systems. Companies including Moonshot, Z.ai and MiniMax are releasing increasingly powerful models at sharply lower cost, challenging long-held assumptions in the West that Chinese developers trail their American peers by months. Moonshot said Kimi K3 is the first open-weight model to approach the 3 trillion-parameter mark and is designed for advanced reasoning, long-horizon coding and knowledge work. The model features a 1 million-token context window, allowing it to process and retain substantially more information than earlier generations in a single prompt. Open-weight models allow users to download, run and customize the underlying systems, unlike proprietary, closed-source models. Kimi K3 "performed competitively with Fable 5 (with fallback) and substantially outperformed Anthropic's Opus 4.8, GPT 5.6 Sol, and GPT 5.5" in terms of GPU kernel optimisation, the company said. The term refers to techniques that maximize AI hardware utilization and minimize latency. The model has also posted strong results in third-party evaluations. Arena.ai ranked Kimi K3 first in a benchmark assessing web interface-building capabilities, while Vals AI placed it second overall behind Fable 5 and ahead of GPT-5.6 Sol. Artificial Analysis said the model delivered performance comparable to OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.8, particularly on tests measuring complex, multi-step tasks. The Moonshot news drove shares of domestic AI competitors Zhipu and Minimax sharply in Hong Kong; just before market close, they were down 27.7 per cent and 16.5 per cent, respectively. Faster release cycles Chinese AI firms are accelerating their model release cycles as the global AI race intensifies. The shift follows the debut of Z.ai's GLM-5.2, which stunned industry observers by scoring near top U.S. closed-source models on benchmark tests, undermining a consensus among Western analysts that Chinese AI models were at least six months behind. Lian Jye Su, chief analyst at Omdia, said Chinese models were gaining traction because they could be deployed far more cheaply than leading U.S. systems. "They can be run at a fraction of the cost that OpenAI charges its clients," he said, but cautioned that Kimi K3's scale didn't "doesn't necessarily mean you have the best performance by default." Kimi K3's size also means few users are likely to host it themselves despite its open-weight release. Ryan Fedasiuk, a fellow at the American Enterprise Institute, said in a LinkedIn post that running a 2.8 trillion-parameter model locally would require hundreds of thousands of dollars of computing equipment. Trillion-parameter systems Parameters are the internal variables a model learns during training and are often used as a rough measure of scale, though not necessarily capability. Before Kimi K3's release, Meituan's LongCat-2.0 and DeepSeek's V4-Pro led China's AI industry with 1.6 trillion total parameters, while several ⁠other domestic rivals have passed the trillion-parameter threshold. But a direct comparison with U.S. frontier models is difficult because companies such as Anthropic and OpenAI do not disclose the parameter counts of systems including Fable, Mythos or GPT-5.5. Moonshot said Kimi K3 incorporates two significant architectural upgrades that improve computing efficiency and enable it to complete long-horizon coding tasks with minimal human supervision. Backed by giants like Alibaba and Tencent, Moonshot has been heavily expanding its capabilities and capital to remain at the forefront of the AI sector. Bloomberg reported last month that the startup was seeking US$2 billion in fresh funding at a valuation of about US$30 billion ahead of a potential Hong Kong listing. --- Reporting by Laurie Chen; Editing by Eduardo Baptista and Shri Navaratnam

Chinese AI firm Moonshot unveils powerful model with capabilities close to Anthropic, OpenAI

Chinese firm Moonshot has unveiled a new open-source AI model with capabilities similar to those of Anthropic and OpenAI - the latest sign that China is catching up to the US in the race to build advanced AI. Dubbed Kimi K3, the large language model was trained on a massive 2.8 trillion parameters - the kernels of data that determine its responses to user questions, according to Moonshot. That would make it one of the largest AI models - if not the largest - ever released. Moonshot, founded by Yang Zhilin, said its tests show the Kimi K3 was outperforming Anthropic's Opus 4.8 model and OpenAI's ChatGPT 5.5 when it comes to most coding tasks, the Financial Times reported. It is still less powerful than Anthropic's cutting-edge Fable model, which was briefly taken offline in June due to concerns that bad actors could misuse it. As an open-source model, Kimi K3 will be available for anyone to download and is expected to be far cheaper to use than the closed-source models offered by CEO Dario Amodei's Anthropic and Sam Altman's OpenAI. For example, Moonshot's K2.6 model was about one-third as expensive as Opus 4.8 to use, according to the FT. Kimi K3 was announced just weeks after China-based Z.AI released GLM-5.2, another open-source model with capabilities at or near those of leading American companies. As The Post has reported, experts have grown concerned that ultra-cheap Chinese models are a major threat to the shaky business models of US labs known for charging top dollar for the "tokens" needed to use their chatbots. If Kimi K3 proves as useful in real-world environments as it did in benchmark testing, it could raise questions about the widely held belief that America has a six to 12 month lead on China for the development of "frontier" AI models. Moonshot could also come under fresh scrutiny as US officials and AI executives warn about China's propensity to rip off technology through the use of unauthorized "distillation" - a technique in which a stronger model is used to train a weaker one. In February, Anthropic publicly called out Moonshot and two other Chinese labs, Deepseek and Minimax, for allegedly using distillation to steal its AI technology. In June, Anthropic sent a letter to Congress detailing evidence that another Chinese tech giant, Alibaba, had distilled its AI models.

China just erased America's AI lead

Why it matters: Kimi K3, a massive new model by Beijing-based Moonshot AI, threatens the foundations of America's AI boom. Its release Thursday dazzled developers, jolted Silicon Valley and reset the AI race overnight. Driving the news: Kimi immediately vaulted into the top tier of global AI, beating Anthropic's Fable 5 and OpenAI's GPT-5.6 Sol in front-end coding tests by AI evaluator Arena. * In Arena's broader text ranking, Kimi finished ahead of Anthropic's Opus 4.8 -- the company's flagship model until Fable 5 arrived in June -- while costing 40% less. * Unlike the premium U.S. models it's challenging, Moonshot plans to release Kimi as an open-weight model on July 27 -- allowing companies and governments to customize and run it on their own systems. The big picture: Even as Chinese open-weight models have gained momentum, U.S. AI leaders and policymakers took comfort in estimates that China remained six to 12 months behind the American frontier. * As recently as April, the U.S. government's AI testing center assessed that Chinese firm DeepSeek's newest model lagged about eight months behind the leading American systems. * Kimi's arrival suggests that cushion may have collapsed far faster than expected. "The entire game has changed. I expect this will trigger some code red for some," AI analyst Kim Isenberg predicted. Between the lines: Kimi does not have to be the world's single best model to upend the market. * For companies, governments and developers, a model that performs near the frontier, costs 40% less and can be customized or run in-house may be the more attractive option. * Its very existence puts pressure on the pricing power of U.S. labs, the enormous valuations built around their technological edge, and the case for spending hundreds of billions of dollars on ever-larger data centers. The other side: America's frontier labs are hardly out of ammunition -- and Kimi may itself reflect the power of U.S. technology. * Anthropic has accused Moonshot and other Chinese labs of industrial-scale "distillation" campaigns, allegedly using millions of exchanges with advanced American models as training data for their own systems. * Chinese companies have obtained restricted Nvidia chips through extensive smuggling networks, despite Washington's efforts to choke off access to the computing power needed to train frontier models. * OpenAI and Anthropic are also racing ahead building newer systems -- including GPT 6 and Claude Opus 5 -- that could restore some distance at the frontier. But the strategic problem remains. Even if U.S. labs pull ahead again, China has shown it can close the gap quickly. What's next: The Trump administration now faces an existential question about how to maintain American AI competitiveness, particularly as calls grow for regulation of frontier models. * Tougher safety rules could slow U.S. labs just as China accelerates; looser oversight could help them move faster while raising the risk of releasing dangerous capabilities. * Restrictions on Chinese models, meanwhile, could protect American companies at home while ceding users abroad. The bottom line: America may still push the frontier forward. It cannot stop the rest of the world from choosing a cheaper alternative.

China's latest open-source AI model further narrows gap with US rivals

Chinese startup Moonshot AI has released Kimi K3, an open-source AI model with specifications on par with Anthropic and OpenAI's newest models. Observers say that China is narrowing the gap between it and the front-runners of the AI industry, and could potentially threaten the US' current market dominance. Moonshot AI launched Kimi K3, its next-generation large language model (LLM), on Friday. The company introduced the model as being able to reach 2.8 trillion parameters, calling it the "world's first open-source model in the 3-trillion-parameter class," stating that it was designed for use in a variety of scenarios, including coding, knowledge work, and reasoning. Kimi K3 is an open-weight model that allows users to download open-source weights, allowing them to use the model for their own servers and even modify it for their specific needs. To the user, this is beneficial in that they do not have to pay fees to access the developer's API. It also reduces their dependence on certain companies. Anthropic and OpenAI and other top AI firms in the US operate closed-weight models that do not allow users access to weights. Moonshot said it plans to release all Kimi K3's weights by July 27. Agencies that assess AI models called Kimi K3 one of the top models currently available. Artificial Analysis, an organization that independently assesses the performance of AI models, gave Kimi K3 a score of 57 points (as of Friday), placing it at No. 3 behind only Claude's Fable 5 (60 points) and OpenAI's GPT-5.6 Sol (59 points). Vals AI, another AI model evaluator, placed Kimi K3 behind Fable 5 but ahead of GPT-5.6 Sol. Moonshot said in its announcement that the model "still trails" flagship Claude and GPT models but has demonstrated "frontier-level performance." While the model features a relatively high token price compared to other open-source Chinese models, it is still much more affordable than Anthropic and OpenAI's cutting-edge models. Reuters said that the model's launch "underscores how quickly China's open AI ecosystem is narrowing the gap with the most advanced US systems." By Kang Jae-gu, staff reporter

China's open-weight Kimi model stuns AI world with frontier-level results

Why it matters: Kimi K3's early performance is fueling awe across the AI world -- and alarm in Silicon Valley and Washington -- as China appears to be rapidly erasing America's lead in advanced AI. Catch up quick: Moonshot says Kimi K3 contains 2.8 trillion total parameters, making it one of the largest open-weight AI models ever released. * It has a 1 million-token context window, allowing it to process enormous amounts of text at once, and can work across text and images. * In blind testing by AI evaluator Arena, developers preferred Kimi over every leading U.S. model for front-end coding -- including Anthropic's Fable 5 and OpenAI's GPT-5.6 Sol. * In Arena's broader text ranking, K3 also outranked the standard version of Anthropic's Opus 4.8 -- a model that sat at the frontier of AI just weeks ago -- and tied Sol. The big picture: Moonshot is offering K3 at prices far below the premium models it is challenging, raising fresh questions about how long U.S. labs can charge top dollar for frontier-level intelligence. What they're saying: "Right now, it's a U.S. versus China question," Mozilla CTO Raffi Krikorian told Axios. * U.S. AI labs are "clearly worried," he said, arguing that their CEOs would have little reason to lobby Washington against open-weight models -- a category led by Chinese companies -- unless they viewed them as a serious competitive threat. Zoom out: Kimi's launch comes just before the 2026 World Artificial Intelligence Conference in Shanghai, where Chinese President Xi Jinping is expected to lay out Beijing's AI priorities. * Moonshot's domestic rival DeepSeek is also expected to release an updated model soon, raising the prospect of another major Chinese breakthrough in quick succession. Yes, but: Kimi has been available for only hours, and early benchmarks and viral demonstrations may overstate how reliably it performs across real-world work. * Moonshot says it will release K3's weights on July 27, meaning developers cannot independently inspect, modify or run the model themselves yet. The bottom line: America's lead in advanced AI may be rapidly shrinking. .

What if China's new AI miracle costs too much?

You know the script. January 2025, an unknown called DeepSeek wipes hundreds of billions of dollars in market value off the board in a single session and convinces the market that you can compete with the Americans for a fraction of the price. Eighteen months later on, here we go again, with blockbuster staging, and perhaps the catalyst Chinese tech needed to bounce back. On July 16, on the eve of Shanghai's major AI expo where Xi Jinping was set to make his debut at the lectern, Moonshot AI rolls out Kimi K3. A model with 2.8 trillion parameters, pitched by Moonshot as the biggest open-source AI model in history. For non-initiates, the weights are the model's brain, everything it learned during training. Moonshot says it will publish them on July 27, and properly equipped companies will then, license in hand, be able to download it, install it on their own servers and customize it to taste, while OpenAI and Anthropic keep theirs under lock and key and sell metered access. Beijing has turned it into a tool of power. That same day, in Shanghai 29 countries signed the founding act of WAICO, a global AI cooperation organization designed for emerging countries, according to Reuters. Behind the machine is a Pink Floyd fan. Yang Zhilin, who studied at Tsinghua and then Carnegie Mellon, named his company in Chinese Yuezhi Anmian as a nod to The Dark Side of the Moon. Almost as strong as the Americans Look at the leaderboards. On the Artificial Analysis Intelligence Index, Kimi K3 posts 57, just behind Anthropic's Claude Fable 5 (60) and OpenAI's GPT-5.6 Sol (59). On WebDev Arena, the ranking dedicated to building web interfaces, it has even temporarily climbed into first place, ahead of Fable 5. Code is already one of the most concrete and most monetized outlets for generative AI. David Sacks, former AI czar and now co-chair of the White House science council, summed up the mood with a spare "this is concerning". Source: Artificial Analysis At Moonshot, they are managing a rich person's problem. Too much demand, not enough servers, new K3 subscriptions have been suspended since Sunday. According to Bloomberg, the company is wrapping up a funding round that values it at $31.5bn, will open talks as soon as August for a final round that could rise to $50bn, and is targeting a Hong Kong listing as early as this year. Its annualized recurring revenue is said to have climbed from more than $200m in April to more than $300m in June. Alibaba and Tencent are among the investors. The low-cost illusion On paper, it looks great. Kimi K3 charges $3 per million input tokens and $15 for output, while GPT-5.6 Sol asks $5 and $30, and Fable 5 a full $10 and $50. Two to three times cheaper than the Americans, the dream. Except the model is a glutton. To reach the same score as Sol, it needs nearly 4 times more tokens. The savings evaporate in real use. Jefferies even estimates an extra cost of about 75% for equal intelligence, while Artificial Analysis, using its own methodology, brings the two models to near parity. The takeaway is simple: the low-cost story is overstated without being a fraud. Moonshot is not hiding it either, its prices jumped 223% versus the previous model, rarely seen from a Chinese player, and K3 now sits among the country's most expensive APIs, according to the broker. Another asterisk is where the performance comes from. Anthropic accuses Moonshot of taking part, alongside DeepSeek and MiniMax, in a broad operation to extract Claude's capabilities: some 24,000 fraudulent accounts and more than 16m exchanges in total, a technique known as distillation. Nothing publicly proves K3 owes its performance to this practice, Jefferies simply notes that its costly operation is consistent with the hypothesis. The star pupil may have glanced at the neighbor's paper. Huawei, the big winner in the standoff That leaves the uncomfortable question: what chips is all this running on? Moonshot will not say, and neither will its Chinese rivals. On the US side, Washington blocks China from getting Nvidia's best chips and allows, under license, only the previous generation. On the Chinese side, Beijing is turning those same chips away at customs, a move widely seen as a way to push its champions toward domestic silicon. The result is that almost nothing moves through official channels, the Commerce Department acknowledged to Congress in mid-July, according to CNBC. Smuggling, meanwhile, is thriving: at least $1bn of banned processors made their way to China in three months in spring 2025, according to the Financial Times. In the middle, Nvidia is paying the bill. Its revenue from customers based in China fell 21% in the fiscal year ended in January, its biggest decline across all regions. The big winner from this two-way blockade is Huawei. Beijing is pushing it as a national champion, and the best students are following. DeepSeek trained part of its latest model on its Ascend chips, and ByteDance, Tencent and Alibaba have moved closer to the group to secure new orders, according to Reuters. At the Shanghai expo, Huawei drew attention with the Atlas 950 SuperPoD, a system that makes more than a thousand chips work as a single machine. But it takes 14 times more chips and 7.4 times more electricity than an Nvidia Vera Rubin NVL72 system, according to Jefferies. China's shortfall in leading-edge lithography is being made up with volume, and that race to scale promises massive investment in the country's foundries, led by SMIC, the broker estimates. For OpenAI and Anthropic, the real danger goes beyond K3's score, and even its bill. China is putting on the table intelligence that is close to the frontier, which a government or company can install on its own servers, bundled with the diplomatic offer that comes with it. It does not need to be better or cheaper, just good enough that the rest of the world stops waiting for permission from Silicon Valley. The shift has already started: Composer, the in-house model from Cursor, the coding tool that SpaceX just bought for $60bn, is built on the open weights of Kimi K2.5, K3's predecessor, by Cursor's own admission. The timing does not help. Both Americans filed confidential IPO paperwork in June: Anthropic could go as soon as the fall, while OpenAI would be looking more at 2027, according to Bloomberg. They will need to convince investors their lead is still worth the price being asked.

What is Kimi K3: Moonshot's frontier AI model that matches Claude Opus 4.8 and GPT 5.5

Moonshot has just launched Kimi K3, which has massive size. It is a 2.8 trillion parameter-sized model, making it the biggest open-weight AI model ever released, and according to Moonshot, it is the world's first "3T-class" open model. The full weights are available starting July 27, 2026, after the technical report. Kimi K3 is designed using two novel architecture components, which are Kimi Delta Attention (KDA) and Attention Residuals (AttnRes), along with a sparser Mixture of Experts architecture that engages only 16 out of 896 experts. Moonshot says that this design results in about 2.5x improved scaling efficiency compared to its previous version, Kimi K2. Also read: What Is Inkling: Thinking Machines Lab's bet on customisable AI On paper, however, the metrics compete. Kimi K3 comes out ahead in the Terminal Bench 2.1 where it receives a score of 88.3 versus Claude Opus 4.8 at 84.6, beating Opus at FrontierSWE (81.2 vs 66.7) and SWE Marathon (42.0 vs 40.0), and comfortably beating GPT 5.5 in agentic benchmarks, including Job Bench and Automation Bench. It loses out to Opus, however, where it scores below in coding and reasoning benchmarks (HLE-Full: 53.3 vs 43.5). This fact is noted explicitly by Moonshot in the blog post itself, which states that K3 "still trails the most powerful proprietary models." K3 makes a point of distinguishing itself with agentic depth. As demonstrated in the examples of case studies provided by Moonshot, K3 autonomously builds a GPU compiler in line with Triton, creates a physical chip design in 48 hours, and replicates a numerical pipeline created in a computational astrophysics paper in two hours, an achievement which Moonshot claims takes a researcher one to two weeks. K3 is also multimodal natively, with a 1-million token context window and vision capabilities integrated into its framework, which is evidenced through demonstrations such as an explanatory video in the style of 3Blue1Brown. Also read: This OpenAI model breached a vending machine to prove a point on AI safety However, there are some caveats that must be mentioned first of all. For instance, according to the limitations section of Moonshot, K3 becomes unstable if an agent doesn't return its entire reasoning process history or if a session is transferred from another model. This looks somewhat strange since a long-term stable reasoning ability was the major feature of the model being discussed. K3 is also noted for being "excessively proactive," which means that it may decide something for its user even without being prompted. Lastly, Moonshot mentions that there is a noticeable gap in the user experience compared to that of Fable 5 and GPT 5.6 Sol. According to the pricing offered by the Kimi API, K3 costs $0.30 per million tokens (cache-hit input), $3.00 (cache-miss input) and $15.00 (output), presenting itself as a cheaper and open alternative, not a leader. K3 is available right now on Kimi.com, Kimi Work, Kimi Code, and the Kimi API. So far, it is fair to say that the main claims made in the framing were proven to be correct thanks to the published numbers. However, one can hardly tell how valid they will remain while using the product.