Alibaba Unveils Qwen 2.5-Max AI Model, Claiming Superiority Over DeepSeek and Other Rivals

Alibaba claims its AI model even better than DeepSeek

Beijing | Chinese tech company Alibaba has released a new version of its artificial intelligence model that it claimed surpassed the highly acclaimed local rival DeepSeek. The unusual timing of the Qwen 2.5-Max's release, on the first day of the Lunar New Year last wek when most Chinese people are off work and with their families, points to the pressure Chinese AI startup DeepSeek's meteoric rise in the past few weeks has placed on not just overseas rivals, but also its domestic competition.

Alibaba Debuts Qwen 2.5-Max AI Model That 'Outperforms' DeepSeek - Decrypt

Ecommerce giant Alibaba has hurriedly released Qwen 2.5-Max as it seeks to remain relevant in a rapidly changing industry. It claims this upgraded model outperforms the likes of ChatGPT, DeepSeek and Meta's Llama "almost across the board." All of this comes as the arrival of DeepSeek, which has been created using cheaper chips and was reportedly built for less than $6 million, continues to wreak havoc among more established competitors. Nvidia saw almost $600 billion wiped off its market capitalization after DeepSeek raced to the top of Apple's App Store -- the biggest single-day drop in Wall Street history. Alibaba's decision to launch Qwen 2.5-Max during the Lunar New Year celebrations underlines how Chinese tech firms are also under pressure. The company's tests claim that it secured a top score of 89.4 in the Arena-Hard benchmark, which compares AI models based on how they respond to human prompts. When it comes to the MMLU-Pro benchmark, which examines whether an AI model can problem solve at college level, Alibaba's testing suggests Qwen 2.5-Max beat DeepSeek and was comparable with ChatGPT. "This endeavor holds the promise of enabling our models to transcend human intelligence, unlocking the potential to explore uncharted territories of knowledge and understanding," developers said on GitHub. President Donald Trump says the powerful performance of Chinese AI models should serve as a "wake-up call" to Silicon Valley -- but told the Congressional Institute he sees the likes of DeepSeek as an opportunity rather than a threat, because it can drive costs down. But there are wider concerns to take into account, especially when it comes to censorship. When asked about the Tiananmen Square massacre, or whether Chinese President Xi Jinping has ever made a mistake, users have noted that DeepSeek refuses to provide answers.

Alibaba Unveils Upgraded AI Model, Claims It Surpasses Rival DeepSeek-V3 -- Update

Chinese e-commerce giant Alibaba released a new version of its artificial intelligence model, claiming it claims surpasses DeepSeek's AI model across various benchmarks. In a statement, Alibaba's cloud unit introduced the latest version of its Qwen large language model, known as Qwen2.5 Max. The company stated that Quen2.5 Max "achieves competitive performance against the top-tier models," referencing OpenAI's GPT-4o, DeepSeek-V3 and Meta's Llama-3.1-405B, based on its compiled comparison using various benchmark tests. Alibaba also announced that Qwen2.5-Max is now available in Qwen Chat, its chatbot-like web interface where users can interact directly with the model. The announcement follows a market frenzy triggered by Chinese AI startup DeepSeek, amid concerns over demand for the most advanced chips and data centers. Just over a week ago, DeepSeek launched its latest open-source AI model, R1. The company claims R1 excels at problem solving, performing on par with OpenAI's GPT-4o reasoning model--but at a fraction of the cost per use. Additionally, DeepSeek's AI assistant app, powered by DeepSeek-V3, recently topped iPhone download rankings in the U.S. These developments have raised security concerns within the industry. On Tuesday, DeepSeek said that it was the target of "large-scale malicious attacks" on its services and said it is temporarily limiting new registrations. However, existing users can continue logging in as usual. Meanwhile, White House Press Secretary Karoline Leavitt said during a Tuesday press briefing that the National Security Council would examine any potential national security implications related to DeepSeek's launch. An OpenAI spokesperson in statement acknowledged that China-based firms and others are "constantly trying to distill the models of leading U.S. AI companies." "As we go forward... it is critically important that we are working closely with the U.S. government to best protect the most capable models from efforts by adversaries and competitors to take U.S. technology," the spokesperson added. News Corp, owner of The Wall Street Journal and Dow Jones Newswires, has a content-licensing partnership with OpenAI.

China's Alibaba enters AI war, claims new model beats DeepSeek, ChatGPT

The company stated that "Qwen 2.5 Max outperforms almost across the board GPT-4o, DeepSeek-V3, and Llama-3.1-405B," highlighting its efficacy against established open-source AI frameworks. This announcement comes after a seismic shift in the AI landscape triggered by DeepSeek's recent releases. Founded only 20 months ago in Hangzhou, DeepSeek has rapidly gained prominence with its AI models, significantly impacting the market dynamics. The introduction of DeepSeek's AI assistant powered by the DeepSeek-V3 on January 10, followed closely by its R1 model on January 20, sent ripples throughout Silicon Valley, leading to noticeable declines in tech stock valuations. DeepSeek claims its V3 model rivals OpenAI's GPT-4o capabilities while offering a more economical choice, resonating with consumers and businesses alike. The company's pricing approach, particularly its V2 model priced at merely 1 yuan ($0.14) per million processed tokens, has activated a price war within China's AI sector. This aggressive pricing strategy has compelled larger companies like Alibaba, Tencent, and Baidu to reconsider their pricing and product development strategies to maintain their foothold in an increasingly competitive market.

Alibaba Unveils Qwen 2.5 AI Model, Claiming It Outperforms DeepSeek And ChatGPT, But It Remains To Be Seen How It Will Be Positioned In The Industry

DeepSeek has shaken the AI industry with its new R1 model, which aims to compete against OpenAI's ChatGPT. While the Chinese AI chatbot is all the hype these days, another major contender has entered the competition, claiming that its AI model is better than that of DeepSeek. Alibaba has officially launched its Qwen 2.5 AI model and claims that it can outperform DeepSeek-V3 along with other leading AI models like ChatGPT. DeepSeek entered the AI industry earlier this week with its new AI model, and it has surpassed ChatGPT to become the most downloaded AI chatbot in various regions, including the United States. With Alibaba stepping into the same field, we can safely presume that China is desperate to take the lead in the category. According to Alibaba's cloud division, "Qwen 2.5-Max surpasses GPT-4o, DeepSeek-V3, and Llama-3.1-405B in almost every aspect," referring to the dominance against Meta and OpenAI (via Reuters). The rapid development of new AI models has intensified the competition in the industry, and the race to become number one has impacted investor confidence in the United States. Companies like NVIDIA have also lost their position in the stock market, making investors reassess their investments in major AI tech giants in the United States. DeepSeek's success in the industry is making rivals step up their game to better compete against the new entrants. Furthermore, existing companies have also released updated AI models to showcase their newer capabilities. For instance, ByteDance unveiled its upgraded AI model and claims it outperforms OpenAI's o1 in AIME, a key benchmark measuring AI comprehension and understanding of complex instructions. With Alibaba's new Qwen 2.5 AI model, we expect the competition to pick up additional heat, which will be evident in the coming months. The previous version of the DeepSeek-V3 also waged a price war for AI models in China. Since DeepSeek-V2 was open source and relatively cheap compared to the competition, with an estimated $0.14 for a million tokens, it led Alibaba to slash its prices by 97 percent on a wide range of AI models. Baidu and Tencent followed suit and cut their prices as well. We have to take note that DeepSeek only has a handful of users compared to Alibaba's hundreds of thousands of employees. Moreover, it only took $5.6 million for DeepSeek to develop the new AI technology against billions of dollars spent in the United States. It remains to be seen how Alibaba's Qwen performs against the likes of DeepSeek and OpenAI and how the company is planning to position itself in the market.

Chinese company Alibaba says its AI model trounces its DeepSeek and OpenAI competitors

A smartphone displays Alibaba's AI-powered assistant Tongyi. (Image credit: Getty Images) Chinese tech company Alibaba has unveiled a new artificial intelligence (AI) model that it claims outperforms its rivals at OpenAI, Meta and DeepSeek. The announcement of the Qwen2.5-Max model yesterday (Jan. 29) is the second major AI announcement from China this week, after DeepSeek's R1 open-weight model took the world by storm following claims that it performs better and is more cost-effective than its American competitors. Now, Alibaba claims that Qwen 2.5-Max, which is also partly open-source, is even more impressive -- surpassing a number of rival models in various tests run by the company. "In benchmark tests such as Arena-Hard, LiveBench, LiveCodeBench, GPQA-Diamond and MMLU-Pro, Qwen2.5-Max is on par with [Anthropic's] Claude-3.5-Sonnet, and almost completely surpasses [OpenAI's] GPT-4o, DeepSeek-V3 and [Meta's] Llama-3.1-405B," Alibaba representatives wrote Jan. 28 in a translated statement on WeChat. Related: DeepSeek stuns tech industry with new AI image generator that beats OpenAI's DALL-E 3 The news comes at an uncertain time for American tech companies. Following DeepSeek's announcement, the AI chatbot quickly overtook ChatGPT to become the most downloaded free app in Apple's U.S. App Store. The company's claims that it achieved better results, while training and running its model at a fraction of the cost, sent shockwaves around the world -- wiping $1 trillion from the valuations of leading tech companies such as Nvidia, whose loss of $589 billion was the biggest one-day market loss in U.S. history. DeepSeek's success has also led to a domestic battle among China's top AI companies, triggering TikTok owner ByteDance to update its Doubao model and likely prompting Alibaba to announce its own new model China's growing competitiveness in AI has become a source of panic for its U.S. counterparts, with OpenAI claiming today (Jan. 29) that DeepSeek plagiarized parts of OpenAI's models to train its own.

Alibaba's Qwen 2.5 AI Claims Edge Over DeepSeek V3

Disclaimer: This content generated by AI & may have errors or hallucinations. Edit before use. Read our Terms of use The cloud division of Chinese tech company Alibaba has released the newest version of its Qwen 2.5 Artificial Intelligence (AI) model that it claims surpasses the DeepSeek V3 in performance, according to a report by Reuters. "Qwen 2.5-Max outperforms ... almost across the board GPT-4o, DeepSeek-V3 and Llama-3.1-405B," Alibaba's cloud unit said in an announcement on its official WeChat account, the report said. The announcement was accompanied by test figures that reportedly showed Qwen 2.5 Max outperforming its competitors. Alibaba, alongside other Chinese tech majors such as Tencent Holdings Ltd and Baidu Inc, has accelerated investments of late into its cloud services segment, and has been reported to be in a competition to recruit AI developers from the country to use its tools. Although Qwen 2.5 delivered promising results, it's important to note that Alibaba tested it against DeepSeek V3, not DeepSeek's latest model, R1. DeepSeek has claimed that R1 performs on par with OpenAI's o1 model in certain AI benchmarks. A A reasoning model, R1 fact-checks itself effectively. While reasoning models take a longer amount of time -- from seconds to minutes -- to arrive at conclusions, they are purportedly more reliable in domains such as science and math. At the same time, several users reported DeepSeek's tendency to censor itself as well. Shortly after R1 hit the market in the US, a huge AI stock sale ensued, resulting in about $593 billion of chipmaker Nvidia's market capitalisation disappearing and its stock prices dropping by 17% by the time US markets closed on Monday. Meanwhile, DeepSeek's AI assistant overtook OpenAI's ChatGPT as the top free application on Apple's App Store. At the time of writing, the same rankings persisted. OpenAI CEO Sam Altman addressed the new AI entrant on his X handle, calling DeepSeek R1 an "impressive model," referring to the output the latter provided. He also added that OpenAI would be bringing forth "much better models," welcoming the competition and saying the company would work to "pull up some releases."

Alibaba announces Qwen 2.5-Max to fight DeepSeek -- what to know

Days after DeepSeek took the internet by storm, Chinese tech company Alibaba announced Qwen 2.5-Max, the latest of its LLM series. The unveiling of this open-source agent can easily be perceived as a direct challenge to DeepSeek and domestic rivals. The release is on the first day of the Lunar New Year when most Chinese people have taken time off work to celebrate and spend time with their families. Alibaba seems to be sending the message that they are hard at work while their competition takes the day off. As we've seen, DeepSeek's emergence has been nothing short of disruptive and Alibaba's new release is sure to cause a stir as well. DeepSeek introduced the R1 model, which delivers performance comparable to leading models like OpenAI's ChatGPT but at a fraction of the development cost. The impact of DeepSeek's innovation was profound, leading to a substantial decline in the market value of major tech companies, including a $593 billion drop for Nvidia -- the largest in U.S. stock market history. The latest from Alibaba comes when the AI industry is witnessing a surge in open-source contributions, allowing researchers and developers worldwide to access and build upon its technology. This move has been lauded for promoting transparency and collaboration in AI development. However, it also puts the power of AI freely into anyone's hands, which could lead to numerous safety and privacy issues. With an emphasis on the model's scalability, Qwen 2.5 has been pre-trained on over 20 trillion tokens and further refined through supervised fine-tuning and reinforcement learning from human feedback. The company has announced the availability of Qwen 2.5's API through Alibaba Cloud, inviting developers and businesses to integrate its advanced capabilities into their applications. Similar to DeepSeek's approach, Alibaba's Qwen 2.5 is built upon the "mix of experts" architecture and aims to not only match but surpass the capabilities of DeepSeek's offerings. This design allows the model to selectively engage different subsets of its parameters, enhancing computational efficiency and enabling the handling of more complex tasks without a proportional increase in resource consumption. It's easy to think that with the rise of these entities aiming to develop more efficient and powerful models, we will continue to see more open-source models unveiled shortly. The ultimate beneficiaries will likely be users who will experience increasingly sophisticated AI applications without the barriers of other chatbots within the category. Alibaba is carefully positioning itself as a key player in democratizing AI. By embracing innovative architectures and open-source principles, Alibaba is responding to the challenges posed by nimble startups like DeepSeek and contributing to the broader advancement of AI technology.

Alibaba launches advanced AI model to rival GPT-4

BEIJING (AFP) - Chinese tech and e-commerce giant Alibaba on Wednesday announced the release of Qwen2.5-Max, an advanced artificial intelligence model that the company says outperforms several leading AI systems in key benchmarks. The launch follows Chinese startup DeepSeek's recent release of models that stunned Silicon Valley and challenged assumptions about US dominance in the booming AI sector. The rapid emergence of consecutive Chinese models will likely intensify concerns in the United States, where companies have invested billions of dollars in AI development that startups in China are matching at significantly lower costs. In a blog post, the Qwen team said their new model outperformed DeepSeek V3 in multiple tests, including code generation and general capabilities, while showing competitive results against industry leaders like OpenAI's GPT-4 and Anthropic's Claude-3.5-Sonnet. The model, trained on over 20 trillion tokens of data, notably was not compared with DeepSeek's R3 model -- the reasoning-focused AI that made waves when it launched through a chatbot on January 20. Qwen2.5-Max is now available to developers through Alibaba Cloud services and can be accessed via Qwen Chat, the company's conversational AI platform. The system offers compatibility with OpenAI's API format, potentially simplifying adoption for organizations already using similar AI services.

Another Chinese AI Company Says It's Beaten OpenAI

Major Chinese tech company Alibaba claims that the latest version of its Qwen AI model has beaten out DeepSeek's V3, the model that flipped Silicon Valley on its head earlier this week by edging out OpenAI. In a statement posted to the Chinese social media platform WeChat, Alibaba said that its Qwen 2.5-Max "demonstrated world-leading model performance in mainstream authoritative benchmarks" and "comprehensively surpasses" OpenAI's GPT-4o, Anthropic's Claude 3.5 Sonnet, and Meta's Llama 3.1. If confirmed, the claim could mark yet another escalation in the race to develop higher performance and more cost-efficient AI models among Chinese competitors, which have thrown Western tech markets into chaos. The news comes after DeepSeek, which was founded in Alibaba's home city of Hangzhou, similarly claimed that it'd figured out how to achieve similar or even better performance than OpenAI's cutting-edge models at a tiny fraction of the cost. The announcement led to a massive scramble, wiping out over $1 trillion worth of market capitalization in former AI frontrunners. Now, Alibaba -- which has far more resources at its disposal than DeepSeek -- has entered the ring, sending even more shockwaves across markets. Alibaba's shares surged nearly two percent following its announcement. In tandem, AI chipmaker Nvidia also slid around five percent Wednesday morning, though whether the drop was in response to Alibaba's unveiling, high-end GPU shortages, or a mix of both, remains unclear. Nvidia was already feeling the hurt, having set a new record for the biggest single-day loss of any company in history. As Reuters points out, it's not just Alibaba and DeepSeek. TikTok owner ByteDance, also based in China, also showed off a new version of its flagship AI model last week that reportedly outperforms OpenAI's o1 AI model in certain benchmarks. As Bloomberg reports, around half a dozen other Chinese AI startups with unicorn valuations are waiting in the wings, hinting at an imminent price war as US-based companies race to compete.

Alibaba Rolls Out AI Model It Says Can Outperform DeepSeek's and OpenAI's

Alibaba's U.S.-listed shares rose 3% in early trading Wednesday following the news. Shares of Meta and OpenAI backer Microsoft (MSFT) were both slightly lower. The Chinese tech giant posted on WeChat that its upgraded Qwen 2.5 Max performed on par with Amazon (AMZN)-backed Anthropic's Claude 3.5 Sonnet, and surpassed OpenAI's GPT-4o, DeepSeek's V3, and Meta's Llama 3.1 in some benchmark tests. The news comes just days after concerns about the surging popularity of an app from Chinese startup DeepSeek, which runs on an AI model it claimed can perform on par with American rivals at a fraction of the cost, sent U.S. tech stocks into a tailspin Monday. After a rebound in Tuesday's session, the tech sector fell back into the red Wednesday, weighing on the major indexes.

Qwen 2.5 Max better than DeepSeek, beats ChatGPT in coding, costs 10x less than Claude 3.5

In case all the buzz about DeepSeek over the past week wasn't enough, Alibaba Cloud launched Qwen 2.5-Max, a state-of-the-art artificial intelligence model designed to outperform industry leaders like OpenAI's GPT-4o and DeepSeek-V3. This release marks a significant milestone in AI development, combining advanced technical capabilities with cost efficiency for enterprise applications. But before we jump to the details, here is how it competes with other well known models: Qwen 2.5-Max is a 72-billion parameter Mixture-of-Experts (MoE) model developed by Alibaba Cloud. Unlike traditional dense models, it uses 64 specialised sub-networks ("experts") that activate dynamically based on the task, reducing computational costs by 30% while maintaining high performance. Also read: DeepSeek AI: How this free LLM is shaking up AI industry The model was pretrained on 20 trillion tokens of data, including academic papers, code repositories, and multilingual web content. It supports text, image, audio, and video processing, with a 128,000-token context window (≈100,000 words), enabling it to analyse lengthy legal contracts or research papers in a single pass. Notably, it can process 20-minute videos, generate SVG code from images, and handle audio inputs in 29 languages, including Mandarin, Arabic, and Hindi. Qwen 2.5-Max leads in reasoning and coding benchmarks but trails Claude 3.5 Sonnet in creative writing tasks. While DeepSeek-V3 is cheaper, Qwen offers better price-to-performance for technical use cases. Unlike open-source rivals, Qwen 2.5-Max is accessible only via Alibaba's API or Qwen Chat interface, limiting third-party customisation. There are certain aspects of Qwen 2.5-Max which not only makes it important but an important player in the AI race. Independent evaluations confirm Qwen 2.5-Max leads in critical benchmarks. On Arena-Hard, a reasoning test requiring multi-step logic, it achieves 89.4% accuracy, surpassing GPT-4o (83.7%) and Claude 3.5 Sonnet (88.1%). Read more: DeepSeek vs OpenAI: Why ChatGPT maker says DeepSeek stole its tech to build rival AI For coding tasks, it scores 92.7% on HumanEval, outperforming GPT-4o's 90.1% and DeepSeek-V3's 88.9%. In scientific reasoning, it achieves 60.1% accuracy on GPQA-Diamond, a benchmark for graduate-level STEM questions, compared to Claude 3.5's 58.3%. These results make it particularly valuable for industries like software development, healthcare diagnostics, and academic research. Priced at $0.38 per million input tokens, Qwen 2.5-Max is 10 times cheaper than GPT-4o ($5/M tokens) and 8 times cheaper than Claude 3.5 Sonnet ($3/M tokens). Read more: DeepSeek is call to action for Indian AI innovation, says Gartner This pricing democratises access to high-performance AI for startups and small businesses, particularly in sectors like finance and education where budget constraints are common. For example, a mid-sized healthcare firm could deploy Qwen 2.5-Max for medical scan analysis at 1/10th the cost of GPT-4o. Released days after DeepSeek's R1 model, Qwen 2.5-Max counters its rival's low-cost disruption. Alibaba's timing intensifies competition in China's AI sector, with ByteDance and Tencent accelerating their own model upgrades. Industry analysts note this positions Alibaba as a key player in the global AI race, particularly for enterprise clients. Despite its strengths, Qwen 2.5-Max has notable limitations. It underperforms Claude 3.5 Sonnet in creative writing tasks, scoring 15% lower on the Creative Writing Benchmark (CWB). Its closed-source nature restricts developer customisation compared to DeepSeek's open models. Additionally, while it processes 128K tokens efficiently, performance degrades slightly beyond 100K tokens in complex tasks. Read more: Deepseek R1 vs Llama 3.2 vs ChatGPT o1: Which AI model wins? Qwen 2.5-Max exemplifies Alibaba's strategy to dominate the enterprise AI market through technical precision and affordability. By outperforming GPT-4o in coding and reasoning at a fraction of the cost, it appeals to businesses seeking advanced AI without prohibitive expenses. While creative applications remain a weakness, its strengths in technical domains position it as a critical tool for industries like healthcare, finance, and software development. As the AI race intensifies, Qwen 2.5-Max underscores China's growing influence in shaping global AI standards.

DeepSeek has rattled the AI industry. Here's a quick look at other Chinese AI models

HONG KONG (AP) -- The Chinese artificial intelligence firm DeepSeek has rattled markets with claims that its latest AI model, R1, performs on a par with those of OpenAI, despite using less advanced computer chips and consuming less energy. DeepSeek's emergence has raised concerns that China may have overtaken the U.S. in the artificial intelligence race despite restrictions on its access to the most advanced chips. It's just one of many Chinese companies working on AI, with a goal of making China the world leader in the field by 2030 and besting the U.S. in their battle for technological supremacy. Like the U.S., China is investing billions into artificial intelligence. Last week, it created a 60 billion yuan ($8.2 billion) AI investment fund, days after the U.S. imposed fresh chip export restrictions. Beijing has also invested heavily in the semiconductor industry to build its capacity to make advanced computer chips, working to overcome limits on its access to those of industry leaders. Companies are offering talent programs and subsidies, and there are plans to open AI academies and introduce AI education into primary and secondary school curriculums. China has established regulations governing AI, addressing safety, privacy and ethics. Its ruling Communist Party also controls the kinds of topics the AI models can tackle: DeepSeek shapes its responses to fit those limits. Here's an overview of some other leading AI models in China. Alibaba Cloud's Qwen-2.5-1M Alibaba Cloud's Qwen-2.5-1M is the e-commerce giant's open-source AI series. It contains large language models that can easily handle extremely long questions, and engage in longer and deeper conversations. Its ability to understand complex tasks such as reasoning, dialogues and comprehending code is improving. Like its rivals, Alibaba Cloud has a chatbot released for public use called Qwen - also known as Tongyi Qianwen in China. Alibaba Cloud's suite of AI models, such as the Qwen2.5 series, has mostly been deployed for developers and business customers such as automakers, banks, video game makers and retailers as part of product development and shaping customer experiences. Baidu's Ernie Bot 4.0 Ernie Bot, developed by Baidu, China's dominant search engine, was the first AI chatbot made publicly available in China. Baidu said it released the model publicly to be able to collect massive real-world human feedback to build its capacity. Ernie Bot 4.0 had more than 300 million users as of June 2024. Similar to OpenAI's ChatGPT, users of Ernie Bot are able to ask it questions and have it generate images based on text prompts. ByteDance's Doubao 1.5 Pro Doubao 1.5 Pro is an AI model released by TikTok's parent company ByteDance last week. Doubao is currently one of the most popular AI chatbots in China, with 60 million monthly active users. ByteDance says the Doubao 1.5 Pro is better than ChatGPT-4o at retaining knowledge, coding, reasoning, and Chinese language processing. According to ByteDance, the model is also cost-efficient and requires lower hardware costs compared to other large language models because Doubao uses a highly-optimized architecture that balances performance with reduced computational demands. Moonshot AI's Kimi k1.5 Moonshot AI is a Beijing-based startup valued at over $3 billion after its latest fundraising round. It says its recently released Kimi k1.5 matches or outperforms the OpenAI o1 model, which is designed to spend more time thinking before it responds and can solve harder and more complex problems. Moonshot claims that Kimi outperforms OpenAI o1 in mathematics, coding, and ability to comprehend both text and visual inputs such as photos and video.

A Guide to China's Leading AI Models

HONG KONG -- The Chinese artificial intelligence firm DeepSeek has rattled markets with claims that its latest AI model, R1, performs on a par with those of OpenAI, despite using less advanced computer chips and consuming less energy. Read More: What to Know About DeepSeek, the Chinese AI Company Causing Stock Market Chaos DeepSeek's emergence has raised concerns that China may have overtaken the U.S. in the artificial intelligence race despite restrictions on its access to the most advanced chips. It's just one of many Chinese companies working on AI, with a goal of making China the world leader in the field by 2030 and besting the U.S. in their battle for technological supremacy. Like the U.S., China is investing billions into artificial intelligence. Last week, it created a 60 billion yuan ($8.2 billion) AI investment fund, days after the U.S. imposed fresh chip export restrictions. Beijing has also invested heavily in the semiconductor industry to build its capacity to make advanced computer chips, working to overcome limits on its access to those of industry leaders. Companies are offering talent programs and subsidies, and there are plans to open AI academies and introduce AI education into primary and secondary school curriculums. China has established regulations governing AI, addressing safety, privacy and ethics. Its ruling Communist Party also controls the kinds of topics the AI models can tackle: DeepSeek shapes its responses to fit those limits. Here's an overview of some other leading AI models in China. Alibaba Cloud's Qwen-2.5-1M is the e-commerce giant's open-source AI series. It contains large language models that can easily handle extremely long questions, and engage in longer and deeper conversations. Its ability to understand complex tasks such as reasoning, dialogues and comprehending code is improving. Like its rivals, Alibaba Cloud has a chatbot released for public use called Qwen - also known as Tongyi Qianwen in China. Alibaba Cloud's suite of AI models, such as the Qwen2.5 series, has mostly been deployed for developers and business customers such as automakers, banks, video game makers and retailers as part of product development and shaping customer experiences. Ernie Bot, developed by Baidu, China's dominant search engine, was the first AI chatbot made publicly available in China. Baidu said it released the model publicly to be able to collect massive real-world human feedback to build its capacity. Ernie Bot 4.0 had more than 300 million users as of June 2024. Similar to OpenAI's ChatGPT, users of Ernie Bot are able to ask it questions and have it generate images based on text prompts. Doubao 1.5 Pro is an AI model released by TikTok's parent company ByteDance last week. Doubao is currently one of the most popular AI chatbots in China, with 60 million monthly active users. ByteDance says the Doubao 1.5 Pro is better than ChatGPT-4o at retaining knowledge, coding, reasoning, and Chinese language processing. According to ByteDance, the model is also cost-efficient and requires lower hardware costs compared to other large language models because Doubao uses a highly-optimized architecture that balances performance with reduced computational demands. Moonshot AI is a Beijing-based startup valued at over $3 billion after its latest fundraising round. It says its recently released Kimi k1.5 matches or outperforms the OpenAI o1 model, which is designed to spend more time thinking before it responds and can solve harder and more complex problems. Moonshot claims that Kimi outperforms OpenAI o1 in mathematics, coding, and ability to comprehend both text and visual inputs such as photos and video.

This Chinese Tech Giant Just Launched an AI Chatbot -- and It's Better Than DeepSeek

Quick Links What Is Qwen Max? Is Qwen Better Than DeepSeek? Summary Qwen Max is a powerful AI model with support for 29 languages and can handle 128,000 tokens. Versions of the AI model like Qwen2.5-Coder-32B excel at coding, while QwQ-32B can think and reason. Qwen outperforms DeepSeek in benchmarks, offering quicker responses and better alignment with human preferences. Before DeepSeek could fade out the headline, another Chinese AI model has come to upseat it. Chinese e-commerce giant Alibaba has announced a new version of its Qwen family of AI models, and there's a lot to like -- it's better than DeepSeek and ChatGPT in some key areas. What Is Qwen Max? Qwen Max is the latest and the most powerful AI model in Alibaba's Qwen AI family. Currently, the following Qwen AI models are available: Qwen2.5-Plus Qwen2.5-Max Qwen2.5-VL-72B-Instruct Qwen2.5-14B-Instruct-1M QVQ-72B-Preview QwQ-32B-Preview Qwen2.5-Coder-32B-Instruct Qwen2.5-Turbo Qwen2.5-72B-Instruct All of the models above are free to use once you've created an account via email, Google, or GitHub. Qwen's AI models are also open-source, meaning you can find them on GitHub or HuggingFace. You can also install them locally on your device (depending on its specs), allowing you to run the AI offline. Qwen2.5-Max is a 72-billion parameter Mixture-of-Experts (MoE) model, supports 29 languages, and is trained on over 20 trillion tokens. It can also handle up to 128,000 tokens in a single conversation, meaning running lengthy documents through the AI will not be an issue. If you're working with data, Qwen can process structured formats like tables, CSVs, and JSON files. As the names suggest, some Qwen models are better at specific tasks. For example, Qwen2.5-Coder-32B-Instruct excels at coding tasks, while QwQ-32B-Preview is capable of thinking and reasoning. Not all of the models can do everything, but most models can handle text prompts, as well as image and video generation. Another rather unique feature is the ability to combine two models together. In my experience, you can get slightly better results when combining two versions into a stronger pairing. For example, pairing Qwen2.5-Max with Qwen2.5-Coder-32B-Instruct helped me generate code with fewer prompts and issues in the output. Qwen's website is the only place to officially access the AI model. You can type in prompts and work with text, but there are image and video generation capabilities as well and in multiple aspect ratios. There's a Web Search feature that has yet to be launched. On the downside, sometimes, Qwen takes a while to process your prompts. So much so that, at first, I thought the website wasn't functioning correctly. I found that the first prompt you send in a conversation can take about 30 seconds to generate a response, after which the responses speed up. Images and videos are generated faster than I expected. They aren't top-of-the-line when it comes to quality or realism, but if you need to generate a quick image in a pinch, they'll do. You can expect a fair amount of random artifacts in most generated media as well. The increased response time could just be because of server load, as was the case with almost every AI chatbot at launch, including DeepSeek and ChatGPT. I did receive errors connecting to Qwen as there were too many requests in the queue from time to time. Is Qwen Better Than DeepSeek? Technically speaking, Qwen is better than DeepSeek across the board. Alibaba's model feels more natural to interact with and runs ever so slightly faster. However, if you were to ignore benchmark results, you'd be hard-pressed to find differences between the two. Qwen's major advantage over DeepSeek is its better alignment with human preferences, making it easier to type in more complicated prompts and get accurate responses without much fine-tuning. Even simple one-liners can generate quite a detailed response with lots of information. DeepSeek, while a terrific AI model, can take a few tries and some prompt engineering before you get your desired results. When it comes to general knowledge and factual accuracy, both models perform similarly, but Qwen does have a slight edge when it comes to factual consistency. One area where DeepSeek is the clear winner, though, is the usage cost. DeepSeek costs $0.25 per million tokens, while Qwen costs $0.38. That said, it's still significantly cheaper than the $5 and $3 rates offered by GPT-4o and Claude 3.5, respectively. Benchmark Comparisons As mentioned before, Qwen outperforms DeepSeek pretty much across the board when it comes to benchmarks. Benchmark Qwen 2.5 Max DeepSeek V3 R1 Arena-Hard 89.4 85.5 MMLU-Pro 76.1 75.9 GPQA-Diamond 60.1 59.1 LiveCodeBench 38.7 37.6 LiveBench 62.2 60.5 The benchmarks clearly show Qwen's better understanding and alignment with human values. Apart from that, in terms of knowledge and reasoning, general knowledge, coding, and overall ability, it's only slightly better than DeepSeek. These two AI models from China have introduced a new benchmark for AI development. There are security and privacy concerns, though, especially considering DeepSeek has already suffered its first data breach. Still, Qwen and DeepSeek's AI models are clearly better than their Western counterparts in terms of performance and have really put the AI world on edge.

Alibaba's Qwen 2.5 AI Model: A New ChatGPT Competitor?

This is a huge claim, and tech enthusiasts are curious to see whether ChatGPT really gets a strong competitor. Qwen 2.5 is the latest artificial intelligence model pre-trained with large-scale multilingual and multimodal data. Along with that, it is programmed in a way that it can adapt to human preferences as people keep using it for different purposes. This LLM model includes vision and audio processing, natural language understanding, text generation, chatbot role-play, and AI tool use. From the very beginning, this AI model is claimed to be released as a direct competitor to . About this, Alibaba's Cloud Unit officially mentioned, "Qwen 2.5-Max outperforms...almost across the board GPT-4o, DeepSeek-V3 and Llama-3.1-405B." ChatGPT and are already established LLMs that, despite some security issues, gained huge success since they were released. So, the competition is going to be tough.

China's emboldened AI industry releases flurry of model updates

Chinese artificial intelligence groups have been rushing out model updates before the lunar new year holiday, as the world wakes up to the sector's major advances led by start-up DeepSeek in the face of US chip restrictions. On Monday, the eve of China's most important annual holiday, the Hangzhou-based company released a new open-source model for image generation, cementing its reputation as the disrupter-in-chief in a field previously dominated by US giants. It came hot on the heels of model releases from tech giant Alibaba and start-ups Moonshot and Zhipu. "This is the equivalent of dropping a massive release on Christmas Eve. We've all been working overtime to get stuff out before the holiday," said one product manager at a large language model start-up. While DeepSeek's achievement has prompted panic in the US about the advances Chinese labs are making on bootstrapped budgets, industry insiders say it is feeding into a newfound "confidence" in China that will spur investment. "DeepSeek has made faster progress than the other Chinese model companies. But this is giving them confidence that they can catch up," said one AI investor in China. DeepSeek has captured the world's attention with a series of model releases that show similar performance to those of US rivals such as OpenAI and Meta, even though it claims to have a fraction of the computing resources and is blocked from acquiring the latest Nvidia processors by US export restrictions. Last week, it released its R1 reasoning model, an advanced model that rivals OpenAI's o1 and can automatically learn and improve itself without human supervision. "DeepSeek has injected a lot of energy into China's AI players and, more broadly, into the global open-source AI community that will use its findings from its R1 paper to make progress on reasoning models," said Wang Tiezhen, an engineer at AI research hub Hugging Face. This week, investors dumped AI-related stocks, with Nvidia losing almost $600bn in market value on Monday. They were reacting to Chinese breakthroughs that show it is possible to build powerful models while pursuing a different strategy to the US one of building ever-larger computing clusters to get ahead in the AI race. On Monday, Alibaba's Qwen released Qwen2.5-1M, a series of new models that are capable of handling longer inputs, an important development that would mean the model could be deployed for AI agent applications with higher memory demands, according to Wang. On the same day, DeepSeek released Janus-Pro, a text-to-image generation model that it claims can surpass state of the art ones from competitors such as OpenAI's Dall-E 3 and Stability AI's Stable Diffusion 3 on some benchmarks. Zhipu, valued at its last funding round in December at $3bn, last week released an update to GLM-PC. The AI agent model is aimed at enterprise customers, enabling computers to automatically complete tasks such as filling out forms or digesting financial reports. While Zhipu has not courted much attention for its LLM development, it has a lead among local AI start-ups in commercialising its technology, with support from local governments and state-owned enterprises that have partnered with the Beijing-based company to deploy its models. Last week, another Beijing-based start-up Moonshot, which owns the popular AI chatbot Kimi, updated its reasoning model to Kimi k1.5, demonstrating strong results compared with established AI models for complex reasoning tasks. The latest release can process texts and images while handling long and complex queries. It is standard practice for Chinese tech companies to release products before the long holiday, with the added benefit that potential customers with lots of free time during the break can test and explore them. Once Chinese AI players return from their break, the race is on to become the leading player developing AI applications for commercial use. "If AI agents can create dramatic commercial value, one or two of the LLM players have a chance to transform into a new generation of software companies," the AI investor said.