Meta launches Muse Spark 1.1 to compete with OpenAI and Anthropic on AI coding

Meta enters the crowded AI coding battle with Muse Spark 1.1

Meta publicly launched a new version of Muse Spark on Thursday, a multimodal AI model designed for agentic coding that aims to compete with similar products offered by OpenAI and Anthropic. Spark 1.1, the first version of which was announced in April, can engage in multi-step reasoning and handle complex processes, manage digital workflows, and deploy new features in enterprise systems, the company says. Meta is a bit behind its competitors here; Anthropic and OpenAI have offered similar models for quite some time. But that doesn't mean Meta's entry into the market isn't a threat. An ongoing source of competitiveness within the AI industry remains the cost of usage, and Meta appears to be offering a competitive rate. Reuters reports that the company will charge $1.25 per million input tokens and $4.25 per million output tokens. That puts it in line (albeit slightly above) Anthropic's Claude Haiku 4.5 and OpenAI's GPT-5.6 Luna. Meta's pitch to users is Spark's ability to handle large agentic workloads, fix bugs, and help with large code migrations -- the kind of automation that enterprises are increasingly turning to AI companies to provide. "Muse Spark 1.1 delivers exceptional performance in personal agentic tasks that require planning and orchestration across a range of external apps and services," the company wrote in a blog post. Meta has released a handful of foundation AI models over the past few years. The Muse Spark release was apparently important enough to compel CEO Mark Zuckerberg to post on the social media platform X for the first time in three years. Zuckerberg's last post was in July 2023, around the time the platform rebranded from Twitter to X. In his post, Zuckerberg called Spark "a strong agentic and coding model at a very low price," noting that the model was "strongest at agentic performance, tool use, and computer use." Zuckerberg also noted that there was "more to come soon" -- implying that the company plans to release additional models. It's been a big week for AI announcements -- particularly for Meta, which also unveiled a new AI image generation model on Tuesday, dubbed Muse Image. Other releases this week have included a new version of Grok from SpaceXAI, and a new family of models from OpenAI, GPT-5.6, that also dropped Thursday. Suffice it to say that the competition within the AI industry is as healthy as ever, and companies that wish to stand out from their peers have their work cut out for them.

Meta Introduces a Big New AI Model for the Agentic Age

Blake has over a decade of experience writing for the web, with a focus on mobile phones, where he covered the smartphone boom of the 2010s and the broader tech scene. When he's not in front of a keyboard, you'll most likely find him playing video games or watching horror movies. Meta on Thursday announced the next version of its multimodal Spark AI model, Muse Spark 1.1, calling it a significant upgrade from its predecessor. The model is built specifically for agentic tasks and has promises major gains in areas such as computer use, coding and multimodal understanding. Spark 1.1 follows the launch of Muse Image, its new image-generation tool, which already has some Instagram users opting out. (You can learn how to do that here.) With the likes of OpenAI and Anthropic storming headlines with new and advanced AI models, Meta's announcement is a sign that the company is jumping back into the AI race after lagging behind for more than a year and undergoing extensive reorganization. The model was released by what Meta calls its Superintelligence Labs, an outfit led by former Scale AI CEO Alexandr Wang. Meta also launched a public preview of its Meta Model API, in which the new Spark 1.1 will be accessible to developers. The model is now available in the "Thinking" mode within the Meta AI app and on meta.ai. Enhanced agentic capabilities Meta says that Spark 1.1 can complete complex projects significantly faster than its predecessor. It's trained to "orchestrate multiagent systems to optimize end-to-end latency," according to the company's blog post. Agentic AI refers to models that can take actual actions on your behalf, rather than just providing answers to a prompt. This multiagent system features a main agent and subagents. The main agent will make a plan and delegate its execution across parallel subagents. Those subagents will stick to their assigned tasks, use available tools and escalate back to the main agent when necessary. Spark 1.1 also actively manages its 1-million-token context window by remembering actions, retrieving information from earlier work and compacting context in a way that keeps the important steps needed for later, according to the blog. That means you should be able to ask it to do more complicated tasks without running into the limits of how much the model can handle at once. Computer use and coding The new model might also be able to take control of your computer for certain tasks, if that's something you're willing to give a try. Meta says its new AI model has received a boost in its computer use workflows across multiple applications. It can maintain context across long sessions and can navigate unfamiliar interfaces with minimal interaction from the user. It can understand when to automate certain tasks and when to navigate through the computer manually. On the coding front, Spark 1.1 promises significant improvements, particularly in tasks involving complex databases. It can diagnose and fix bugs, add new features in enterprise-grade systems and perform large code migrations. Truly multimodal Spark 1.1's multimodal capabilities allow it to see and hear to complete tasks. This isn't unlike existing AI models we've seen today, like Gemini Live's camera experience. It can ingest imagery and audio, and provide "ultra-descriptive" captions from the sources. Combining this with its agentic computer use abilities, it can interpret what's on the screen and take action to perform tasks. Safety Meta is touting its extensive safety evaluations for the model, which follows the Advanced AI Scaling Framework. The model is said to show a strong resistance to jailbreaks, prompt injection and other common attacks.

Meta says its new AI model is ready to compete on coding

After reentering the AI race with its first in-house Muse Spark model in April, Meta is now opening up the doors to developers with a new model that can plug into AI coding software with the new Meta Model API. Meta says that Muse Spark 1.1 is a "step-change" from the first generation, with improvements based on feedback from developers. The company says it's capable of more advanced coding, including detection and fixing of complex bugs; better supports end-to-end agentic workflows across a range of apps, including multi-agent systems; and has native multimodal perception across images, videos, and documents. The Muse Spark 1.1 launch follows this week's launch of Muse Image, an image generation model that's proved controversial for its ability to incorporate other users' Instagram content into its generations. It's part of Meta's race to justify the billions it's spent on catching up in the AI competition and attempting to achieve parity with companies like OpenAI, Google, and Anthropic, especially after a slew of high-profile hires and a company restructuring last year. The new 1.1 model is available now in Thinking mode through the Meta AI app and Meta AI website. It will also be accessible through a new Meta Model API, available from today in public preview for US developers. Meta is including $20 worth of free credits with every new Meta Model API account. Muse Spark was initially only available directly through Meta AI, before eventually powering the chatbots inside Instagram and WhatsApp, and the latest Meta smart glasses.

Meta launches low-cost Muse Spark 1.1 as enterprise AI spending comes under scrutiny

Meta says the model delivers competitive performance against OpenAI, Anthropic, and Google offerings while costing a fraction as much to run. Meta has unveiled Muse Spark 1.1, saying the frontier AI model rivals leading LLMs on coding, computer use, and agentic AI benchmarks while undercutting OpenAI and Anthropic on API pricing, potentially lowering the cost of deploying AI agents in enterprises. The latest model, which was teased last week, matched or was competitive with leading models, such as Claude Opus 4.8, Gemini 3.1 Pro, and GPT 5.5, across several agentic AI, coding, and computer-use benchmarks, including SWE-bench Verified, Terminal-bench, BrowseComp, SpreadsheetBench, and OSWorld, Meta wrote in a blog post. Muse Spark 1.1, which is currently in public preview and available via the Meta Model API, will cost $1.25 per million input tokens and $4.25 per million output tokens, the company noted.

Meta debuts Muse Spark 1.1 with preview open to developers

July 9 (Reuters) - Meta Platforms (META.O), opens new tab on Thursday released long-awaited developer access to its Muse Spark AI model alongside an upgraded version, pitting it directly against the business models of Anthropic and OpenAI in charging for use of its AI. The social media giant touted Muse Spark 1.1 as its most capable model for real-world coding and agentic tasks, part of a broader mission the company is pitching of delivering "personal superintelligence." Meta said the upgraded model can write and debug code, use software and external tools, understand text, images and video, and carry out complex multi-step tasks ⁠with less human intervention. In April, Meta debuted Muse Spark, the first text and reasoning AI model from the superintelligence team it assembled last year to close the gap with rivals in the heated competition for AI supremacy. Meta was testing the Application Programming Interface with partners in a private preview during its launch. The API is a key element for AI systems, acting like a digital bridge for developers that allows them to use the model's capabilities in their own software systems. Developers in the United States can now access Muse ⁠Spark in public preview on Meta Model API, letting them test prompts, compare outputs and prototype integrations. Those who sign up for the API receive $20 in free credits to test the model before switching to pay-as-you-go pricing. The access is priced at $1.25 per million input tokens and $4.25 per million ⁠output tokens, above OpenAI's entry-level GPT‑5 mini and Anthropic's low-cost Claude Haiku 4.5, but below Anthropic's higher-end Claude Sonnet 4.6 model. The new model is now available in Thinking mode in the Meta ⁠AI app and on the website. It is also expected to replace existing Llama models powering chatbots on WhatsApp, Instagram, Facebook and Meta's collection of smart glasses. The release ⁠follows a company announcement on Tuesday expanding generative AI tools across its apps by rolling out Muse Image, its first image-generation model from Meta Superintelligence Labs. Reporting by Harshita Mary Varghese in Bengaluru and Katie Paul in New York; Editing by Pooja Desai Our Standards: The Thomson Reuters Trust Principles., opens new tab

Meta jumps into AI coding market in effort to chase Anthropic and OpenAI

Three months after unveiling its first artificial intelligence model under the leadership of AI chief Alexandr Wang, Meta is rolling out a major update as it attempts to compete with OpenAI and Anthropic in critical areas of the market. Muse Spark 1.1, which Meta introduced on Thursday, represents its "strongest model for agentic and coding work yet," Wang said in an interview with CNBC. The initial Muse Spark model released in April was only available to "select partners" who could access the technology via a "private API preview." Meta is making the new model's API available through a developer portal as part of a public preview, where users will be able to sign up and see instructions for integration. A Meta spokesperson said some early partners can already access the API, and new users "will be able to add themselves to a waitlist and be added from there over time." For now, Meta said it's limiting API access to its own properties rather than making it available on third-party platforms like the popular OpenRouter marketplace. "This is going to be served on top of the computer infrastructure that we've built," Wang said. It's Meta's second notable rollout for the Muse family this week. On Tuesday, Meta released Muse Image, originally code-named Mango, a model for creating images, as the company seeks to attract creators and advertisers to its offerings. Meta CEO Mark Zuckerberg is coming under pressure from Wall Street to show a return on the company's massive and growing investment in AI infrastructure and development. While it's spending at the rate of its hyperscaler peers, Meta doesn't have a cloud infrastructure business (though it plans to start one), and it's failed to keep up with OpenAI, Anthropic and Google in developing popular models and AI applications.

Meta Launches New A.I. Model as Global Technology Race Heats Up

For the first time, the company will offer a paid version of the service, a departure from its longtime philosophy of giving its A.I. away for free. Meta on Thursday unveiled a new version of its flagship artificial intelligence model, Muse Spark, in an effort to close the gap with its rivals in the global race to develop the technology. For the first time, the company will offer a paid version of the service, a departure from its longtime philosophy of giving its A.I. away for free. The paid product opens up a new revenue stream for Meta, which is planning to spend tens of billions of dollars on A.I. this year. The new model is better at coding than Meta's previous version of the technology, which the company released in April. On tests that measure writing, reasoning, coding and other tasks, Muse Spark performed at or near the same levels as leading models from Anthropic, OpenAI, Google and xAI, according to data shared by the company. Meta's product would also be cheaper than other leading models, the company said. Muse Spark costs customers about a fourth what Anthropic's leading model, Fable, does. Muse Spark is the first A.I. model developed by Meta Superintelligence Labs, the company's A.I. division that Mark Zuckerberg, Meta's chief executive, revamped last year. Along with spending billions on new A.I. researchers to develop models like Muse Spark, Mr. Zuckerberg has pledged to spend $600 billion to build new data centers in the coming years. The new A.I. division is led by Alexandr Wang, Meta's 29-year-old chief A.I. officer and the former chief executive of the start-up ScaleAI. Meta released its first image generator developed under Mr. Wang, called Muse Image, on Tuesday. Meta also plans to release a video generator called Muse Video and an even more powerful A.I. model internally called Watermelon in the coming months. Meta is facing fierce competition in the A.I. race from rival tech firms. Elon Musk's A.I. company, xAI, on Wednesday released a new A.I. model, and OpenAI released its latest model on Thursday. Mr. Zuckerberg promoted Meta's new model on X on Thursday, calling it "strongest at agentic performance, tool use, and computer use." "More to come soon," he added.

Meta thinks its newer, cheaper AI is the best way to beat ChatGPT and Gemini

The model supports text, images, video, audio, and PDFs, plus a massive 1 million-token context window for complex tasks. The AI race has mostly been about who can develop the most powerful model. Meta now wants to turn the conversation to who can offer one for the cheapest price. Meta is taking a swipe at rivals with the launch of Muse Spark 1.1, its first paid AI API for developers. As noted by Bloomberg, the company will charge about 25% of what the top AI providers do to access Muse Spark. This will make it a much cheaper option than OpenAI and Anthropic models. So far, the company has built its ecosystem largely on open-weight AI models and free access to developers. Muse Spark 1.1 introduces a new pricing model: developers are able to use the model for free up to a token limit and then pay for additional usage via the new Meta Model API. It's the first time Meta has seriously charged companies to access one of its flagship AI models. The strategy comes amid Meta's efforts to justify the massive amounts it is investing in AI. The company has committed hundreds of billions of dollars to new data centers, chips, and talent, as investors have wondered how those investments will ultimately pay off. Meta keeps its free consumer AI experiences while charging for API access, opening a new revenue stream. Low prices don't mean much if performance isn't competitive, and Meta is betting Muse Spark 1.1 does both. The model is a significant step forward in coding, tool use, multimodal reasoning, and agentic capabilities -- artificial intelligence systems that can perform complex, multi-step tasks with minimal user input, the company said. It can handle text, images, video, audio, and PDFs; has a context window of up to one million tokens; and is built to drive everything from coding assistants to enterprise AI agents. Meta says Muse Spark 1.1 is competitive with today's AI models on several benchmarks, including coding and reasoning tasks. The same model will power Meta AI across Facebook, Instagram, WhatsApp, its standalone app, and Ray-Ban Meta smart glasses, as the company previously said when it introduced Muse Spark. The bigger question is whether low pricing is enough to disrupt a market where OpenAI, Anthropic, and Google have built up strong developer ecosystems. Meta is betting it can get developers to take its AI stack seriously by offering competitive performance at lower prices -- and make its enormous AI spending a viable business in the process.

Meta's AI chief says new Muse Spark update will sharpen coding, agentic AI

Alexandr Wang said the upcoming Muse Spark update will significantly improve coding and agentic capabilities, while analysts say the model could give enterprises a lower-cost alternative to OpenAI and Anthropic. Meta is set to launch a new Muse Spark model with stronger coding and agentic capabilities, as Chief AI Officer Alexandr Wang touted the update as a step toward closing the gap with rival AI platforms and expanding the company's enterprise AI ambitions. "..Our next Muse Spark update is coming soon. Big improvements in coding and agentic capabilities to be more competitive with other leading models," Wang wrote in a post on X in an attempt to clarify CEO Mark Zuckerberg's comments about the slow progress in AI agent development made during a company townhall. In the same townhall, Wang said that the next Muse Spark update, codenamed Watermelon, which uses far more compute than its predecessor, has already caught up with OpenAI's flagship GPT 5.5 model, according to a Business Insider report that cited anonymous sources.

Meta updates its Spark model, releases developer version

Driving the news: Muse Spark 1.1 promises improvements in coding and handling longer tasks. * Muse Spark, the first large language model developed under Alexandr Wang, already powers a wide range of tasks across Facebook, Instagram, WhatsApp and Meta AI. * With this version, developers can also tap the model via an API (Application Programming Interface). Meta says it believes its model is priced attractively given its capabilities, charging customers $1.25 per million input tokens and $4.25 per million output tokens. (That's cheaper than the new Grok 4.5, and also less than Anthropic's Opus model.) What they're saying: Wang, who was brought in to reboot Meta's AI effort, told Axios that improving Muse Spark's performance in coding and agentic tasks was a key focus with this release. * "These were priority areas for us," he said. Meta is also using the model to power the thinking mode in the Meta AI app and website. The big picture: Meta is trying to close the gap between it and AI leaders including OpenAI and Anthropic. The company is promising a significantly more powerful version of Spark later this year that uses substantially more computing power. * Earlier this week, Meta debuted Muse Image, a picture-generating model that has been added to Meta AI and powers new image editing features in Instagram. * This is not the big leap that Meta hopes to make with its next big release. Code-named Watermelon, that model is still in training and uses vastly more compute and should be released later this year. What's next: Longer term, Wang said he is working toward a vision where a more agentic Meta AI could handle tasks like planning a party or a vacation and take more direct actions on its own. Wang said Meta's strength is its billions of users and how much the company knows about them.

Meta's latest AI model is Muse Spark 1.1 and it can run your computer for you

AI assistants have gotten really good at answering questions and walking us through complicated tasks. But the next wave of AI is aiming for something much bigger: doing those tasks for us. That's the idea behind Meta's new Muse Spark 1.1. Instead of simply telling you which buttons to click, the model is built to interact with your computer on your behalf. Whether it's searching across multiple websites, filling out forms, or switching between apps, Meta says Muse Spark 1.1 can navigate software much like a person would, choosing the fastest way to finish the job. It's a notable shift from purely conversational AI to AI designed to take action. The era of point-and-click AI is here The biggest upgrade is that Muse Spark 1.1 can take control of the task itself. The company says the model is designed to work across desktop apps, web browsers, and even mobile interfaces. Depending on what's faster, it can either interact with apps the way you would -- clicking buttons and navigating menus -- or automate the process by generating scripts behind the scenes. Picture planning dinner with a group of friends. Instead of bouncing between Google Maps, restaurant websites, and the group chat, an AI agent could compare places, check opening hours, collect the details, and present you with the best options. You still make the final call, but the tedious legwork is largely taken care of. Recommended Videos That's where the AI race is heading. Companies are no longer competing to build the most talkative chatbot -- they're trying to build assistants that can actually get work done. Muse Spark 1.1 also comes with a sizable memory upgrade, supporting a context window of up to one million tokens. That means it can keep track of lengthy conversations, massive documents, and complex projects without constantly needing a refresher. Can it really handle the messy stuff? Like every major AI release, Muse Spark 1.1 arrives with a stack of benchmark wins. Meta says the model holds its own against some of the biggest names in AI across coding and agent-focused tests, and even comes out ahead in several. Those results are certainly eye-catching, but benchmark charts have never been the final word. What really matters is how the model behaves once it leaves the controlled environment of a test. Can it navigate dozens of websites without losing track of what it's doing? Will it adapt if a webpage suddenly changes? Can developers trust it to work through large codebases without needing constant supervision? Those are the questions that will determine whether Muse Spark 1.1 is genuinely useful or just another impressive demo. Meta is giving developers the chance to find out for themselves. Muse Spark 1.1 is now available in public preview via the new Meta Model API, enabling developers to build applications and run the model through real-world workloads. Whether Muse Spark 1.1 becomes the AI assistant people actually depend on remains to be seen. But one trend is impossible to ignore: the race is no longer about building chatbots that write better responses. It's about building AI that can roll up its sleeves, take over the repetitive work, and let you focus on everything else.

Meta releases latest update of AI model Muse Spark as tech giant accelerates AI push under Alexandr Wang | Fortune

Meta released a new version of Muse Spark on Thursday, claiming that the AI model surpasses the capabilities of prior products from OpenAI, Anthropic, and Google. Dubbed Muse Spark 1.1, the new model is proficient across a range of tasks, including coding, video captioning, and reasoning, Meta said in a blog post. The update to Muse Spark beats out Google's latest release of Gemini in benchmarks measuring coding and reasoning capabilities, according to the tech giant. And Meta's new AI entry surpasses older versions of OpenAI and Anthropic's models on some verticals, the company said. "Our focus is on delivering strong agentic and multimodal models at very low cost," said Mark Zuckerberg, Meta's cofounder and CEO, in a post on X. "More to come soon." The tech giant did not say how its new AI product compares to OpenAI and Anthropic's most recent model releases. On one coding metric, for example, the model still lags behind the flagship entries -- Anthropic's Mythos 5 and Fable 5, and OpenAI's GPT-5.6 -- according to a leaderboard ranking performances on an open-source benchmark. In the past, the company has also confronted accusations that it manipulated results on tests for a prior AI model release. "We've also heard claims that we trained on test sets -- that's simply not true and we would never do that," a former Meta AI executive said on X in response to allegations in April 2025. Superintelligence Labs The release of Muse Spark 1.1 comes just three months after Meta unveiled its first significant entry into the AI race under a reorganized unit headed by Alexandr Wang, Meta's first-ever chief AI officer. Originally committed to building open-source models, the tech giant pivoted after it spent $14.3 billion in 2025 to acquire a 49% non-voting stake in Scale AI, Wang's AI data startup that skyrocketed in prominence amid the AI boom. Wang reorganized the company's AI efforts into what it now calls Meta Superintelligence Labs. The organizational changes weren't without drama. Staffers felt whiplashed as Wang and Zuckerberg rebuilt the company's AI team from the ground up. Zuckerberg also created an Applied AI unit in March, which pulled in engineers and developers to help on what some saw as mind-numbing data collection tasks. Still, Meta's efforts to catch up to OpenAI and Anthropic appear to be accelerating. Muse Spark now powers Meta's AI assistant in its Meta AI app and developers can access the newest version of the model via an API. On Tuesday, Meta also released Muse Image and Muse Video, the first models released from Superintelligence Labs focused on image and video generation. The new visual AI models produced impressive results. But Meta's decision to allow users to apply AI-powered image generation and editing effects to photos that other people had posted to their public-facing Instagram accounts without requiring any explicit permission to do so, sparked outrage from many Instagram users. Muse Spark 1.1 is now in public preview. Meta did not say when it would fully release the model, but Zuckerberg did state that the model will be lower cost with "aggressive pricing" compared to competing products from OpenAI and Anthropic.

Meta enters pay-to-use AI market with Muse Spark 1.1

Meta describes its latest iteration as a 'significant upgrade' over the last model. Following a series open source models over recent years, Meta begins selling access to AI with its latest launch, Muse Spark 1.1. The model marks the second product under the Alexandr Wang-run Meta Superintelligence Labs (MSL) and comes three months after the flagship series Muse Spark was first unveiled. Meta described its latest iteration as a "significant upgrade" over the last model, while Mark Zuckerberg told Bloomberg that this could be the first time "Meta's models are better than all of the Google models". MSL chief Wang told CNBC that the new product represents Meta's "strongest model for agentic and coding work yet". Muse Spark 1.1 will include a new paid tier for developers, which according to Zuckerberg, will be "among the most affordable options" in the market. "The pricing is going to be very aggressive and attractive. "We think that there's a real ability to be able to offer frontier or very high-level intelligence at a much more affordable cost," he told Bloomberg. Meta's API pricing is roughly one-fourth of what OpenAI and Anthropic charge for their top models. The new model can take over a wide array of tasks across the device, including assigning tasks to subagents. It can also remember actions and retrieve information from much earlier work. Zuckerberg said that the model has "state-of-the-art or very close to it" agentic reasoning and tool use. Muse Spark 1.1 showcases significant leaps over Muse Spark in a number of benchmarks, while performing at par or a few points ahead similar models from OpenAI, Google or Anthropic. It marks 72 in vibe coding, leaps ahead of Muse Spark, which reached just under 20. Meta is carving itself another revenue stream, after planning to spend around $145bn in capital expenditure this year. A majority of that spend is focused on its AI ventures. The company is also reportedly planning to enter the cloud computing market by building a business that would sell access to AI computing power and models to external customers. Earlier this year, it signed a $21bn deal with CoreWeave to access the company's AI cloud capacity until December 2032. Despite reporting a revenue growth of 33pc over the quarter past, morale at Meta is seemingly at a low, with employees expressing frustration at the mounting work pressure created by the company's major AI plans. MSL, however, is "doing better than we expected", Zuckerberg told Bloomberg. The company has a new model under development codenamed 'Watermelon', which the Meta CEO said could help further its positioning. 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.

Meta launches flagship Muse Spark 1.1 model with multi-agent upgrades

Meta launches flagship Muse Spark 1.1 model with multi-agent upgrades Meta Platforms Inc. today launched a new flagship large language model optimized to power multi-agent automation workflows. Muse Spark 1.1 is available in the company's Meta AI chatbot service and via an application programming interface. The Meta Model API, as it's aptly called, will enable developers to embed the LLM in their custom software. The service is in public preview. Multi-agent automation workflows usually comprise a main agent that develops a plan for how to perform tasks and subagents that carry out its instructions. The main agent generates its plan at the start of a new project. According to Meta, Muse Spark 1.1 can detect and respond to mid-task developments that make it necessary to change the project plan. Tasks that require multiple agents to automate usually comprise a large number of steps. As a result, agents generate a significant amount of data while performing those steps. If the dataset exceeds the underlying LLM's context limit, some of the information has to be discarded, which often comes at the expense of output quality. Muse Spark 1.1 addresses the challenge with a context compaction mechanism. According to Meta, it compresses the data generated by AI agents in a way that saves the most important details. That enables Muse Spark 1.1 to retrieve "information from much earlier work" when it needs to carry over data from one sub-task to another. The model has a context window of 1 million tokens. Muse Spark 1.1's context compaction and multi-agent features make it adept at coding tasks. In one internal test, Meta engineers had the model generate a chat app based on prompts. Muse Spark 1.1 took screenshots of the program's interface, spotted several technical issues, identified the code snippets that caused the issues and fixed them. Muse Spark 1.1 scored 72.2 on an AI programming benchmark called Vibe Code Bench v1.1, which puts it more than 50 points ahead of Meta's previous flagship LLM. It achieved a nearly 18% higher score on a second test called SWE-Atlas Codebase QnA. The features that make Muse Spark 1.1 adapt at generating code also enable it to tackle other use cases. It can generate e-commerce listings based on a product video, place restaurant orders on the user's behalf and perform other multi-step tasks. The Meta Model API through which developers can access Muse Spark 1.1 presumably runs on the Facebook parent's own infrastructure. Reuters reported today that the company plans to boost its data center capacity to 14 gigawatts next year. The plan is said to revolve around Iris, an internally developed AI chip set to enter mass production in September. The report indicates that Iris is likely the MTIA400, a custom AI processor Meta previewed in March. It includes 51% more HBM memory bandwidth than the company's previous-generation silicon. Additionally, the chip supports enhanced versions of MX8 and MX4, two data formats that AI models use to store their information. Meta says that MTIA400 is 400% faster than its predecessor. The company's custom silicon gives it the option to follow up the Meta Model API with additional enterprise offerings. For example, the Facebook parent could sell on-premises AI appliances that combine its MTIA chips with Muse Spark 1.1. Other hyperscalers are also moving to make their custom AI accelerators available to third party data center operators.

Meta launches Muse Spark 1.1 for agentic AI coding

Meta launched Muse Spark 1.1 on Thursday, a multimodal AI model focused on agentic coding that aims to compete with similar offerings from OpenAI and Anthropic. The model, first announced in April, can perform multistep reasoning, manage digital workflows, and deploy new features in enterprise systems, according to the company. Meta's entry into the field occurs as it trails competitors which have had similar models in the market. However, the company asserts that its competitive usage costs will enhance its standing. Reuters reported that Meta will charge $1.25 per million input tokens and $4.25 per million output tokens, slightly above the rates for Anthropic's Claude Haiku 4.5 and OpenAI's GPT-5.6 Luna. Meta promotes Muse Spark 1.1's capacity to handle large agentic workloads, fix bugs, and assist with large code migrations. The company stated in a blog post that the model provides "exceptional performance in personal agentic tasks that require planning and orchestration across a range of external apps and services." CEO Mark Zuckerberg marked the launch by posting on X for the first time in three years, having last engaged with the platform in July 2023, when it rebranded from Twitter. In his post, Zuckerberg described Spark as "a strong agentic and coding model at a very low price," emphasizing its performance in agentic tasks and tool use. Zuckerberg indicated that the company intends to release additional models in the near future, stating that there would be "more to come soon." This week also saw Meta unveil a new AI image-generation model called Muse Image, alongside a series of announcements from other AI companies, including a new version of Grok from SpaceXAI and OpenAI's GPT-5.6, both released on Thursday. The competitive landscape within the AI industry continues to intensify, as firms strive to differentiate themselves amidst growing demand.

Meta to release new AI model with advanced coding capabilities 'soon'

Meta to release new AI model with advanced coding capabilities 'soon' Meta Platforms Inc. is gearing up to release a new version of its flagship Muse Spark artificial intelligence model. Alexandr Wang, the company's Chief AI Officer, wrote on X today that the update will roll out "soon." The announcement came a few hours after Business Insider reported that the new algorithm is competitive with GPT-5.5 across several "closely followed" AI benchmarks. The publication's sources didn't name the benchmarks. Practically all frontier model announcements include results from SWE-Bench Pro, a benchmark used to evaluate AI systems' coding capabilities. The original version of Muse Spark scored 52.5% on the test. GPT-5.5, the OpenAI Group PBC model that Meta's new algorithm reportedly matches, reached 58.6%. GPT-5.5 also outperformed Muse Spark on another popular coding benchmark called Terminal-Bench 2.0. Wang stated on X that Meta's upcoming model is significantly more adept at coding than Muse Spark. He added that it's also better at powering AI agents. Muse Spark includes a "contemplating mode" that uses AI agents to improve the quality of prompt responses. During its internal testing, Meta had the model complete a benchmark called HLE with and without the feature. Muse Spark scored 8% higher when contemplating mode was enabled. An X user asked Wang when Meta would launch a model that can match the coding capabilities of Anthropic PBC's Claude Opus 4.8. The executive responded by stating it will happen "pretty soon." Claude Opus 4.8 can perform some coding tasks significantly better than both Muse Spark and GPT-5.5. It scored 69.2% on SWE-Bench Pro, or 10.2% higher than OpenAI's flagship model. However, it fell behind GPT-5.5 on Terminal-Bench 2.0. The improved output quality of Meta's new model reportedly comes at the expense of increased infrastructure use. According to Business Insider, the algorithm uses an "order of magnitude" more computing capacity than Muse Spark. Meta's push to improve the coding capabilities of its AI models suggests it may be planning to make them accessible to external developers. On Wednesday, Bloomberg reported that the Facebook parent is considering launching an AI infrastructure service. Executives are reportedly debating whether Meta should offer raw computing capacity or hosted AI models. Taking on existing AI development tools such as Claude Code will require Meta to build more than just coding-optimized AI models. Claude Code offers integrations with popular developer tools, a desktop app and customization features. Users can also configure it to repeat a task at specific time intervals. Coding is not the only use case to which Meta could apply its planned Muse Spark successor. Anthropic offers Claude Code alongside Claude Cowork, a productivity tool geared towards non-technical professionals. The latter offering includes several vertical-specific feature bundles geared towards industries such as the healthcare and financial sectors.

Mark Zuckerberg Is Spending Hundreds of Billions on AI. His New Strategy in the Race Against OpenAI Is Surprisingly Simple.

The AI race has produced massive data centers, expensive chips and billion-dollar talent wars. Mark Zuckerberg's latest weapon is much simpler: keep the prices down. Meta unveiled Muse Spark 1.1, a new AI model with a paid tier for developers, creating a new revenue stream as the company races against OpenAI, Anthropic and Google. Zuckerberg says the new Meta Model API will cost roughly 25% of what competing AI models charge. "The pricing from some of the other labs is very extreme," Zuckerberg told Bloomberg. It's a big change for Meta, which has spent years releasing its AI models for free. The company has made cuts to invest heavily in AI infrastructure as Zuckerberg pushes toward his goal of creating superintelligence. Zuckerberg says building its own models is critical to creating AI products that can reach billions of users, rather than relying on competitors to power Meta's future.

Meta Unveils Muse Spark 1.1 AI Model With 1 Million-Token Context Window

Developers are using it for software development, AI research workflows Meta Platforms on Thursday announced Muse Spark 1.1 as its latest multimodal reasoning model designed for agentic AI tasks. The company claims it brings significant improvements over its predecessor when it comes to tool use, coding, computer interaction, and multimodal reasoning. Developed by Meta Superintelligence Labs, the AI model supports a context window of up to one million tokens. Alongside, Meta also released a public preview of the new Meta Model API for developers. Muse Spark 1.1 Available to Developers via Meta Model API The company says that Muse Spark 1.1 is available to developers via the new Meta Model API that is now in public preview. It is claimed to be designed to perform complex agentic tasks that require planning, reasoning, and coordination across multiple applications and external services. The AI model can generalise to new native tools, Model Context Protocol (MCP) servers, and custom skills without any additional training, Meta said in a blog post. The company highlighted the improvements in task orchestration, with Muse Spark 1.1 being capable of serving as a primary agent that gathers context, formulates plans, and delegates work to multiple subagents running in parallel. Compared to the original Muse Spark model, its architecture is claimed to deliver faster completion of projects while reducing end-to-end latency. Meta is also touting Muse Spark's one million-token context window, which helps it manage long-term memory, improve information retrieval from previous conversations, and its ability to compress the context while retaining key details for future use. Other areas of improvement include computer-use capabilities and coding. As per the company, Muse Spark 1.1 can navigate desktop apps, interact with unfamiliar software interfaces, and take a call on when tasks should be automated using scripts. It also promises better performance during enterprise-scale software engineering tasks like debugging large codebases, implementing new features, executing large-scale code migrations, and building complete web apps. Meta claimed that its developers are already internally using the model for software development and AI research workflows. Benchmark results shared by Meta show its competitiveness against other AI models across several industry evaluations. The model is claimed to have achieved an MCP Atlas score of 88.1, JobBench score of 54.7, Humanity's Last Exam score of 62.1, and Finance Agent v2 score of 57.2. On coding benchmarks, it recorded 80.0 on Terminal-Bench 2.1 and 53.3 on DeepSWE 1.1. On the security front, Meta claims to have conducted safety evaluations under its Advanced AI Scaling Framework before deployment. It is said to have remained within acceptable threshold margins across chemical, biological, cybersecurity, and loss-of-control risk categories. Further, the AI model is claimed to have demonstrated improved resistance to jailbreak attempts, prompt injection attacks, and hallucination compared to the original model.

Meta debuts Muse Spark 1.1 with preview open to developers

The social media giant touted Muse Spark 1.1 as its most capable model for real-world coding and agentic tasks, part of a broader ⁠mission the ⁠company is pitching of delivering "personal superintelligence." Meta Platforms on Thursday released long-awaited developer access to its Muse Spark AI model alongside an upgraded version, pitting it directly against the business models of Anthropic and OpenAI in charging for use of its AI. The social media giant touted Muse Spark 1.1 as its most capable model for real-world coding and agentic tasks, part of a broader ⁠mission the ⁠company is pitching of delivering "personal superintelligence." Meta said the upgraded model can write and debug code, use software and external tools, understand text, images and video, and carry out complex multi-step tasks with less human intervention. In April, Meta debuted Muse Spark, the first text and reasoning AI model from the superintelligence team it assembled last year to close the gap with rivals in the heated ⁠competition for AI supremacy. Meta was testing the Application Programming Interface with partners in a private preview during its launch. The API is a ⁠key element for AI systems, acting like a digital bridge for developers that allows them to use the model's capabilities in their own software systems. Developers in the United States can now access Muse Spark in public preview on Meta Model API, letting them test prompts, compare outputs and prototype integrations. Those who sign up for the API receive $20 in free credits to test the model before switching to pay-as-you-go pricing. The access is priced at $1.25 per million input tokens and $4.25 per million output tokens, above OpenAI's entry-level GPT‑5 mini and Anthropic's low-cost Claude Haiku 4.5, but below Anthropic's ⁠higher-end Claude Sonnet 4.6 model. The new model is now available in Thinking mode in the Meta AI app and on the website. It is also expected to replace existing Llama models powering chatbots on WhatsApp, Instagram, Facebook and Meta's collection of smart glasses. The release follows a company announcement on Tuesday expanding generative AI tools across its apps by rolling out Muse Image, its first image-generation model from Meta Superintelligence Labs.

After SpaceX, Meta Now Launches Muse Spark 1.1 Model, to Begin AI Chip Production in Sept

This is the second company in two days that has suddenly joined the AI coding battle where the likes of Anthropic and OpenAI are ruling the roost Two days and two AI laggards have suddenly gotten into overdrive with their latest AI models with one of them also announcing that they'll also have their own custom AI-chips within the next couple of months. Close on the heels of SpaceX announcing a new AI model comparable to those of established players, Meta Platforms has now announced the Muse Spark 1.1. As with the Grok 4.5 model, the latest from the Meta stable also claims its latest offering would be great for AI coding with the ability to handle large agentic workloads, fix bugs, and assist with code migrations - those that the automation junkies within enterprises have promised for months but delivered little. While we speculated yesterday around why Elon Musk's SpaceX suddenly woke up and landed a new frontier model, it looks like similar reasons might be behind the measures taken by Meta and Mark Zuckerberg. He actually told Bloomberg that he aims to use Meta's pricing as an "aggressive and attractive" option in an ecosystem where tokenmaxxing is causing concerns. For now, we may've to wait for some time before to check whether Meta's claims of delivering performance comparable to AI models of Google, Anthropic and OpenAI. But, there is no doubt that Zuckerberg is hoping to spend considerable money to subsidise usage as an effort to pull market share from those who're leading the AI race currently. The Bloomberg report said Meta's pricing would be around a quarter of what the likes of Anthropic and OpenAI are charging for their top models. Whether this price war would achieve the desired results will depend on what the company is pitching itself as. Meanwhile, Reuters reported that Meta would also start making the latest versions of its AI-chips in September, with design support from Broadcom and manufacturing by TSMC. In addition, the company would also buy RAM from Samsung, storage from Sandisk and fibreoptic equipment from Sumitomo Electric. The sole purpose being reduction in GPU costs. Watch out Nvidia! The report made these revelations based on some internal memos that Reuters got access to. Of course, Meta's chip dreams aren't new as they'd detailed four new chips back in March that was to be developed under an accelerator program. They've a modular approach to chip design in view of the belief that the needs will change as AI evolves. Coming to the latest foundational model release, we cannot but wonder how Meta would make money with it. Maybe, that's not their immediate intention (as with other AI labs) and the punt now is on how enterprises respond to the high AI costs. So, undercutting prices might be one way to steal a few customers and then keep them hooked. Watch out for future price increases! Maybe, they consider that the chip-making exercise would reduce their reliance on Nvidia and help save some of the costs on that front. From Meta's point of view, they intend using the new chips for training models, for ranking and recommendation algorithms, larger AI workloads, and of course the most importance inference demand. Meta has made chips since 2023.

Zuckerberg Breaks 3-Year Silence on X to Launch Meta's New AI Model

Zuckerberg chose to share this news on X, a platform he had avoided for years. This choice alone has people talking. Meta Model API, the company's first proprietary API in the US, lets users employ Muse Spark 1.1 within their own apps. These moves together show Meta wants to capture a larger share of the developer market, a space where OpenAI and Anthropic currently lead. said the new model can handle complex tasks. It works with third-party tools and holds a large amount of information at once, with a context window of up to 1 million tokens. In plain terms, the model can keep track of several details during a single task. It can also run several AI agents simultaneously. The model works on desktop, mobile, and browser platforms. said it can write code, spot and fix bugs, automate tasks, and work with text, images, and video. The goal is to help businesses automate workflows with less human effort. On the pricing side, developers can start with free credits and pay for usage after the credits run out. Reports say the cost is $1.25 per million input tokens and $4.25 per million output tokens. Meta confirmed the model will show up in its own apps too. A new 'Thinking' mode is coming soon to the Meta AI app and website.

Meta superintelligence labs drops Muse Spark 1.1 By Investing.com

Investing.com -- Meta Platforms' specialized AI division, Meta Superintelligence Labs, announced Thursday the release of Muse Spark 1.1. The updated multimodal reasoning model is specifically engineered to handle autonomous "agentic" workflows, boasting a massive one-million-token context window alongside major upgrades to coding and direct computer control. The release marks a significant push by Meta to lead the enterprise AI market. Alongside the consumer-facing rollout, the company launched a public preview of its Meta Model API, opening up the Muse architecture to outside developers for the first time. The defining architecture of Muse Spark 1.1 is its ability to operate natively within multi-agent networks. According to Meta, the model can function simultaneously as a "primary agent" and a "subagent," allowing it to break down large tasks and delegate them across parallel subagents to drastically reduce processing latency. The model is currently live in "Thinking" mode for consumers on the Meta AI app and meta.ai. Meta is positioning Muse Spark 1.1 as a direct competitor to frontier engineering models, claiming substantial improvements in parsing massive, enterprise-scale codebases. Internal testing on the Meta Internal Coding Bench indicates the model outpaces its predecessor and runs competitively against unnamed industry alternatives. The company states it can autonomously diagnose bugs, implement new features into legacy enterprise code, and execute large-scale code migrations. Beyond writing code, Muse Spark 1.1 introduces practical "computer use" features. The model can automate complex workflows across multiple desktop applications by writing its own automation scripts or interacting directly with user interfaces when text-based scripting isn't viable. A trio of prominent early API partners -- Replit, Cline, and Box -- headlined the announcement, signaling strong early adoption in the developer tooling space. "The combination of a million-token context window, deep multimodal support, and advanced coding capabilities fundamentally changes how we think about scale," said Amjad Masad, CEO of Replit. Saoud Rizwan, CEO of Cline, pointed to the model's tool-use flexibility and competitive pricing for high-volume developer workloads, while Yashodha Bhavnani, VP of AI Products at Box, noted that Spark 1.1 performed head-to-head with top-tier frontier models on Box's internal enterprise evaluation sets. Addressing ongoing industry concerns regarding autonomous AI systems, Meta noted that the model underwent strict safety auditing under its Advanced AI Scaling Framework. The evaluations specifically tested for catastrophic thresholds, including chemical and biological risks, cybersecurity vulnerabilities, and "loss of control" scenarios. Meta stated the model operates safely within its established margins, adding that a comprehensive evaluation report will be published shortly. The rollout of Muse Spark 1.1 coincides with the general release of Muse Image, Meta's latest standalone image-generation model.

Meta debuts Muse Spark 1.1 with preview open to developers

July 9 (Reuters) - Meta Platforms on Thursday released long-awaited developer access to its Muse Spark AI model alongside an upgraded version, pitting it directly against the business models of Anthropic and OpenAI in charging for use of its AI. The social media giant touted Muse Spark 1.1 as its most capable model for real-world coding and agentic tasks, part of a broader mission the company is pitching of delivering "personal superintelligence." Meta said the upgraded model can write and debug code, use software and external tools, understand text, images and video, and carry out complex multi-step tasks with less human intervention. In April, Meta debuted Muse Spark, the first text and reasoning AI model from the superintelligence team it assembled last year to close the gap with rivals in the heated competition for AI supremacy. Meta was testing the Application Programming Interface with partners in a private preview during its launch. The API is a key element for AI systems, acting like a digital bridge for developers that allows them to use the model's capabilities in their own software systems. Developers in the United States can now access Muse Spark in public preview on Meta Model API, letting them test prompts, compare outputs and prototype integrations. Those who sign up for the API receive $20 in free credits to test the model before switching to pay-as-you-go pricing. The access is priced at $1.25 per million input tokens and $4.25 per million output tokens, above OpenAI's entry-level GPT-5 mini and Anthropic's low-cost Claude Haiku 4.5, but below Anthropic's higher-end Claude Sonnet 4.6 model. The new model is now available in Thinking mode in the Meta AI app and on the website. It is also expected to replace existing Llama models powering chatbots on WhatsApp, Instagram, Facebook and Meta's collection of smart glasses. The release follows a company announcement on Tuesday expanding generative AI tools across its apps by rolling out Muse Image, its first image-generation model from Meta Superintelligence Labs. (Reporting by Harshita Mary Varghese in Bengaluru and Katie Paul in New York; Editing by Pooja Desai)

Mark Zuckerberg returns to X after 3 years to launch Meta's Muse Spark 1.1 AI model, developer API

Meta has launched a paid API for developers in the US, positioning itself against OpenAI and Anthropic in the enterprise AI market. Meta CEO Mark Zuckerberg has posted on X (formerly Twitter) for the first time in more than three years to announce the latest artificial intelligence model, Muse Spark 1.1, along with a new API that gives developers access to the model. Taking to X, Zuckerberg described Muse Spark 1.1 as the company's latest model that focuses on coding, software development and AI agents. He said that the model is designed to handle complex tasks, work with external tools and process long-running workflows using a context window of up to one million tokens. According to him, the model can coordinate multiple AI agents simultaneously and interact with desktop, mobile and browser interfaces. Along with the announcement, Meta opened a public preview of its new Meta Model API for developers in the US. The API allows the developers to integrate Muse Spark 1.1 into their own applications. Also read: OpenAI introduces ChatGPT Work, an AI agent powered by GPT 5.6 and Codex: Here is what it can do As per the company, Muse Spark 1.1 is capable of writing and debugging code, automating software workflows and processing text, images and videos. Meta says the model is designed for tasks that require minimal human intervention, making it suitable for enterprise automation and AI-powered applications. The developers signing up for the platform will receive free credits before moving to a usage-based pricing model. Reports indicate the service is priced at $1.25 per million input tokens and $4.25 per million output tokens. Meta is also bringing Muse Spark 1.1 to its consumer-facing AI products through a new "Thinking" mode in the Meta AI app and website. Over time, the company is expected to expand the model across its ecosystem, including WhatsApp, Instagram, Facebook and its AI-powered smart glasses.