Meta unveils Muse Spark AI model as Superintelligence Labs makes its debut

Meta's Superintelligence Lab unveils its first public model, Muse Spark

Meta on Wednesday announced Spark, the first AI model in the Muse family that it says represents "a ground-up overhaul of our AI efforts." Muse Spark is the first release of Meta's Superintelligence Labs, formed a little less than a year ago with the grandiose goal of "deliver[ing] on the promise of personal superintelligence for everyone." The release represents a clean break from Meta's previous work on the open source Llama model family, which has received a middling reaction both from users and on independent LLM rankings. And while Spark will be a proprietary model, Meta founder and CEO Mark Zuckerberg said in a post on Threads that the Muse family will "includ[e] new open source models" in the future. Meta said that Muse Spark will take advantage of content posted across platforms like Instagram, Facebook, and Threads, much as xAI's Grok is integrated with content posted on X. Currently, this means Muse Spark can link to public posts related to a location or trending topic that you ask about, for instance. In the future, Meta says this will expand to "new features that cite recommendations and content people share" and "Reels, photos, and posts woven directly into your answers, with credit back to the content creators." Contemplation and compression In a somewhat technical blog post accompanying the Spark announcement, Meta includes the by now routine laundry list of AI benchmarks, in which Muse Spark's standard thinking mode ranks comparable or better than competing models from OpenAI, Anthropic, Google, and xAI. But that post also sheepishly acknowledges that "we continue to invest in areas with current performance gaps, such as long-horizon agentic systems and coding workflows." In that same post, Meta also touts "Contemplating" mode, which it says will be "rolling out gradually" and which can "orchestrate multiple agents that reason in parallel." By using up to 16 agents thinking in concert simultaneously, Meta says that Contemplating mode "enables superior performance with comparable latency." That "superior performance" includes a reported high water mark of 58.4 on Humanity's Last Exam (with the use of external tools), according to Meta. And while previous Llama models faced criticism for not taking advantage of reinforcement learning, Meta says Muse Spark shows "smooth predictable gains" after additional RL steps after pretraining, "improving model reliability without compromising reasoning diversity." That reinforcement learning system also makes use of "thinking time penalties," which Meta says balance the need to "maximize correctness" with optimizing the number of tokens used. In testing on the AIME 2025 benchmark, Meta says it saw a "phase transition" where the model started compressing equally accurate reasoning into "significantly fewer tokens." After that compression, subsequent trained models slowly increased the token usage again to achieve even higher accuracy in less overall time than the previous uncompressed versions. The release of Muse Spark comes alongside an update to Meta's Advanced AI Scaling Framework, which the company says now covers a broader range of potential AI risks. The company says that the model "falls within safe margins across all frontier risk categories we measured," but says that more details will only be available in an upcoming Safety & Preparedness Report. Muse Spark is available now in the Meta AI app and via the meta.ai website, as well as a private preview API for "select partners." Meta says the model will be available via WhatsApp, Instagram, Facebook, Messenger, and AI glasses "in the coming weeks."

Meta AI app climbs to No. 5 on the App Store after Muse Spark launch | TechCrunch

Meta's AI app has seen a sizable jump in installs following Wednesday's launch of the company's newest AI model, Muse Spark -- its first model release under Alexandr Wang, the head of Meta's Superintelligence Labs who was recruited from Scale AI last year to overhaul the social giant's AI efforts. According to new data from market intelligence provider Appfigures, consumer demand for the Meta AI app has pushed the app up from No. 57 right before the launch of Muse Spark on Wednesday to No. 5 on the U.S App Store on Thursday -- a move suggesting a flood of new installs. Meta says its new AI model, which is available on both the web and mobile, is a significant upgrade over its earlier Llama 4 models. It's also the company's latest attempt to catch up to rivals like OpenAI and Anthropic, an effort that's already cost Meta billions in recruiting AI talent, in addition to its $14.3 billion-dollar investment in Scale AI. Currently, Muse Spark accepts multimodal input, including voice, text, and images, and has been designed to perform well on a number of tasks, like helping people learn about their health and reasoning through complex questions in areas like science and math. It can also aid in visual coding, letting users create websites and mini-games from prompts. Plus, Meta AI is able to launch multiple subagents to handle users' questions, the company said. The model will roll out to other platforms, including WhatsApp, Instagram, Facebook, Messenger, and Meta's AI glasses, in the weeks ahead. Alongside the model's launch, the Meta AI mobile app and website were upgraded with a new look and feel, and now allow users to switch between modes depending on the task. Despite the recent growth, Meta AI's app for now still lags behind the AI chatbots from other top model makers, including OpenAI's ChatGPT (No. 1), Anthropic's Claude (No. 2), and Google's Gemini (No. 3). Wang pointed to the new high rank in a post on X earlier Thursday, and he noted that the app is "still growing." Appfigures data indicates that Meta AI's app has been installed a total of 60.5 million times worldwide across both the App Store and Google Play, with 25 million of those downloads occurring just this year. Over the past five months, Meta AI app downloads have increased by 138% when compared with the first five months of the app's availability. India is now Meta AI's top market by downloads, followed by the U.S., Brazil, Pakistan, and Mexico, according to Appfigures.

Meta debuts the Muse Spark model in a 'ground-up overhaul' of its AI | TechCrunch

Meta released an AI model on Wednesday called Muse Spark, which marks its "first step" toward an "overhaul of [its] AI efforts." Muse Spark is the inaugural model to come out of Meta Superintelligence Labs, which was created last year because CEO Mark Zuckerberg was reportedly unhappy with the progress of Meta and its Llama models and how they lagged behind OpenAI's ChatGPT and Anthropic's Claude. Meta recruited former Scale AI co-founder and CEO Alexandr Wang to lead Meta Superintelligence Labs and invested $14.3 billion in the data labeling company for a 49% stake. Now, it's time for Zuckerberg to see if his reconfigured AI team can woo users. Muse Spark, which is now available on the web and the Meta AI app, is expected to improve over time. The company plans to roll out a "Contemplating" mode, which allows it to tackle more complex problems. Meta's model uses multiple AI agents at once to work on the same problem, which it says will generate faster results for its Contemplating mode. "To spend more test-time reasoning without drastically increasing latency, we can scale the number of parallel agents that collaborate to solve hard problems," the company wrote. Meta's competitors have historically placed these more capable models behind a paywall. It's unclear if Meta will follow the same strategy. The company has jumped on one AI industry trend, though. Meta said in its blog post that Muse Spark could be applied to help users with health questions, something that competitors are also working on as well. Meta's push into health -- and even the more basic function of logging into Muse Spark -- could raise privacy concerns. Muse Spark users will need to log in with an existing Meta account such as Facebook or Instagram in order to use it. Meta doesn't explicitly say that personal information from a Facebook or Instagram account will be used by the AI. But it is likely considering that Meta generally trains on public user data and the company has positioned Muse Spark as a personal superintelligence product. Meta also said that Muse Spark performs especially well with visual STEM questions which can lead to "interactive experiences like creating fun minigames or troubleshooting your home appliances." Aside from its investment in ScaleAI and hiring of Wang, Zuckerberg's company has recruited researchers from OpenAI, Anthropic, and Google. The upshot: if Meta's going to be a real competitor in the AI industry, it's now or never. "Looking ahead, we plan to release increasingly advanced models that push the frontier of intelligence and capabilities, including new open source models," Zuckerberg wrote on Threads. "We are building products that don't just answer your questions but act as agents that do things for you."

Meta Unveils New AI Model Developed by Costly, New Superintelligence Labs

Expertise I have more than 35 years' experience in journalism in the heart of Silicon Valley. Meta unveiled its latest artificial intelligence model on Wednesday, its first since assembling a costly team to help the company better compete with its AI rivals. The new model, known as Muse Spark, was created by Meta Superintelligence Labs, a team of expensive AI leaders and developers assembled last year from other AI companies and led by Alexandr Wang, the co-founder and CEO of Scale AI. Wang joined the Meta AI team after Meta agreed to make a $14.3 billion investment in Scale AI. The model, known internally as Avacodo, now powers the Meta AI app and the Meta AI website and will soon power Meta apps WhatsApp, Instagram, Facebook, Messenger and AI glasses. Meta called the new model "an early data point on our trajectory," adding that it has larger models in development. "This initial model is small and fast by design, yet capable enough to reason through complex questions in science, math, and health," Meta said in its announcement. "It is a powerful foundation, and the next generation is already in development." Meta has been investing increasingly large sums to compete in the ever-evolving AI landscape, creating a "superintelligence team" that works toward achieving artificial general intelligence, or AGI. The release of Meta Muse comes after the company's disappointing release of the Llama models last year. Meta has bet big on AI to catch up and keep pace with other AI builders. Google leapfrogged its rivals in November with its Gemini 3 model, showcasing its impressive coding and research abilities. OpenAI was quick to follow with updates to GPT-5.

Meta is reentering the AI race with a new model called Muse Spark

Meta Superintelligence Labs is launching its first model since Mark Zuckerberg spent billions overhauling the company's AI efforts. Called Muse Spark, the model now powers the Meta AI app and the Meta AI website in the US, per the company's announcement. In the coming weeks, Meta says, it will appear in WhatsApp, Instagram, Facebook, Messenger, and Meta's smart glasses, as well as roll out in other countries. Like Google Gemini, which easily integrates into Google's product suite, Meta touts Muse Spark as "purpose-built for Meta's products." The model, the first in a new series, will also be available to some of Meta's partners in private preview" via the API. The company promises the ability to run multiple AI sub-agents to handle queries better and faster, as well as support for multimodal input that includes both text and images. The latter is particularly relevant to Meta's AI-powered camera glasses, which it's bet on as the (latest) future of computing. It lets users toggle between a faster "Instant" mode and a "Thinking" mode that's supposed to deliver more thoroughly reasoned results, similar to options like Microsoft's Think Deeper. Meta also highlighted that Muse Spark can answer "complex questions in science, math, and health." Health-focused AI chatbots have been a controversial topic in recent months, as they handle sensitive personal data and can propagate misinformation. Meta said that Muse Spark's multimodal perception is "especially valuable for health" and can "navigate health questions with more detailed responses, including some questions involving images and charts." Meta may be looking to compete with OpenAI's ChatGPT Health and Anthropic's Claude for Healthcare, which both debuted in January. In its announcement, it showed its chatbot estimating a calorie count for a meal -- a popular, but often hit-or-miss, use of AI tech. In the future, Meta hopes the model will power new features "that cite recommendations and content people share across Instagram, Facebook, and Threads." The company also said that it has larger models in development and hopes to open-source future versions. It describes Muse Spark as an "early data point" on the trajectory of its new Muse series. The Muse series is set to be Meta's second major foray into powerful AI, following its Llama models. Zuckerberg revamped the company's AI program after the delayed and disappointing release of Llama 4 in 2025.

Meta's new model is as open as Zuckerberg's private school

You were the chosen one! It was said that you would destroy the proprietary models, not join them! Nearly two years after extolling the virtues of open source AI, Meta CEO Mark Zuckerberg is singing a different tune. On Wednesday, the social media magnate unveiled its first new model developed by its Superintelligence team. But this is no Llama, and you can't download its weights. The model, dubbed Muse Spark, is proprietary with access limited to Meta's AI portal or API access for those lucky enough to get an invite. Yep, this thing's locked down tighter than Zuck's private school. Detailed in a blog post, Meta described the model as the "first step on our scaling ladder and the first product of a ground-up overhaul of our AI efforts." Meta's decision not to release Muse Spark's weights comes in stark contrast to Zuckerberg's earlier position that "open source AI represents the world's best shot at harnessing this technology to create the greatest economic opportunity and security for everyone." Those comments came forth in a 2,000-plus-word manifesto titled "Open Source AI is the Path Forward," in which Zuckerberg waxed poetic on the merits of open source AI. "If we were the only company using Llama, this ecosystem wouldn't develop and we'd fare no better than the closed variants of Unix," Zuckerberg wrote, drawing comparisons to the rise of the Linux operating system. "Opening Llama doesn't undercut our revenue, sustainability, or ability to invest in research like it does for closed providers," he argued, emphasizing that Meta's business model didn't depend on selling access to its models, either. That was the case until about a year later, when Meta launched its Llama API inference service, alongside its Llama 4 family of models. But never fear, Zuckerberg hasn't given up on open models entirely. "Looking ahead, we plan to release increasingly advanced models that push the frontier of intelligence and capabilities, including new open source models," he wrote in a Threads post. This kind of dual-tracking is common. In fact, Google routinely releases small open weights models derived from its larger proprietary Gemini models, with its Gemma 4 family being the latest example. OpenAI has made similar moves with gpt-oss, though it remains to be seen if that was a one-off or not. But if Zuckerberg actually believed any of what he wrote in 2024, why bother with a closed model in the first place? Perhaps because even though Meta hyped Llama 4's multimodal and agentic capabilities, the model never lived up to expectations, with Meta ultimately abandoning development of its largest variant, codenamed Behemoth, which would have weighed in at 2 trillion parameters. The flop was apparently embarrassing enough that Meta started over from scratch, paying top dollar to woo top AI software engineers and executives, including Alexandr Wang, who now leads Meta Superintelligence Labs. So it's not open source, you can't download it, but is it any good? It's a good question, because if it's not, Meta's shareholders might start to wonder why the company is so keen to light $135 billion on fire. If Meta is to be believed, Muse Spark is a big improvement over Llama 4. The model boasts performance matching and in many cases besting the top models from OpenAI, Anthropic, and Google. But before you read too far into these benchmark numbers, remember that they're coming from a company that not that long ago was accused of pulling a bait-and-switch in order to make Llama 4 look better. Having said that, this time around Meta had the foresight to share its test methodology. Compared to Llama 4, Meta claims Muse Spark was also more efficient to train, proving that "we can reach the same capabilities with an order of magnitude less compute than our previous model." Meta didn't go into detail about the model's underlying architecture, describing it as a "natively multimodal reasoning model with tool-use, visual chain of thought, and multi-agent orchestration" capabilities. Meta has also introduced what it calls a "contemplating mode," which orchestrates multiple reasoning agents working in parallel to compete with frontier models like Gemini Deep Think and GPT Pro. However, it doesn't appear that this function is generally available on day one. "Muse Spark is available now, and Contemplating mode will be rolling out gradually in meta.ai." Spark is just the first in a new line of Muse models, with larger variants already in the works - and unlike Behemoth, we might actually get to see them. ®

Meta Debuts First AI Model From Prized Superintelligence Group

Meta Platforms Inc. debuted its latest artificial intelligence model Wednesday -- its first since Chief Executive Officer Mark Zuckerberg embarked on a multibillion-dollar overhaul of the company's AI organization to keep pace with rivals. The highly anticipated model, known as Muse Spark, was created by Meta Superintelligence Labs, the new team of expensive AI researchers helmed by Chief AI Officer Alexandr Wang. Muse Spark will power the Meta AI chatbot, and in a pivot from the company's prior open-source strategy, is a closed model, meaning its design and code won't be made public. Meta shares jumped 6% in New York following the announcement. The model is the first major test for Zuckerberg's new AI lab, known as MSL. The Facebook founder recruited Wang as part of a $14 billion investment into Scale AI last year after a series of setbacks left him frustrated that the company wasn't keeping pace with competitors like OpenAI, Anthropic PBC, and Alphabet Inc.'s Google. Zuckerberg has since spent billions on AI talent and committed tens of billions more on infrastructure like data centers to catch up. Meta has tried to keep its AI division nimble, giving researchers autonomy and minimizing its typical management-heavy organizational structure, a company executive said. Wang has around 100 direct reports, said the executive, who asked not to be named discussing internal matters. The executive acknowledged that Muse Spark wasn't as capable in some areas as OpenAI's ChatGPT, Anthropic's Claude or Google's Gemini, but said that it was early in the company's execution. The model is "an early data point on our trajectory," with several larger models in development, Meta said in a blog post. Muse Spark, known internally as Avocado during development, was built over nine months and is seen by executives as a refresh of Meta's AI strategy, which was previously focused on its open-source Llama models. Wang is a proponent of closed models, and while Meta still plans to build open-source ones in the future, it is also considering selling API access to Muse Spark, according to the company executive. The Meta AI chatbot will remain free for users, but the company is considering implementing subscription fees in the future, the executive said. The name Muse Spark hints at the broader pipeline of AI offerings to come: The model is "sparking," or jump-starting, MSL's efforts, the executive said, describing it as small and fast. It is the first in a series of expected "Muse" models -- an homage to Zuckerberg's interest in Greek and Roman mythology. Get the Tech Newsletter bundle. Get the Tech Newsletter bundle. Get the Tech Newsletter bundle. Bloomberg's subscriber-only tech newsletters, and full access to all the articles they feature. Bloomberg's subscriber-only tech newsletters, and full access to all the articles they feature. Bloomberg's subscriber-only tech newsletters, and full access to all the articles they feature. Plus Signed UpPlus Sign UpPlus Sign Up By continuing, I agree to the Privacy Policy and Terms of Service. Muse Spark was trained using several third-party open-source models, including Qwen from the Chinese tech giant Alibaba Group Holding Ltd., as well as ones from OpenAI and Google. Using a Chinese model runs counter to the stance taken by US policymakers and tech executives, who have raised concerns about China's position in a fast-moving AI race -- citing national security risks. Some of the largest American AI companies have tried to crack down on Chinese models similarly trained on US technology. "Like others across the industry, Meta uses techniques like distillation with strict safeguards in place to learn from openly available AI models and improve our own," said a Meta spokesperson. Like its competitors, Muse Spark offers several levels of reasoning: "Instant" mode, "Thinking" mode and "Contemplating" mode -- the last providing research-grade responses. The Meta executive said the model is good at answering questions about science, health and math, but weaker at coding. The model will be used to improve Meta's fleet of apps, from Instagram to Facebook to WhatsApp. It is already powering an early-stage shopping agent that aims to help users better search for products like clothing or furniture. The agent cannot yet make purchases on a user's behalf, and is still in early stages, the executive said.

Meta unveils first AI model from costly superintelligence team

April 8 (Reuters) - Meta Platforms (META.O), opens new tab on Wednesday unveiled Muse Spark, the first artificial intelligence model from a costly team it assembled last year to catch up with rivals in the AI race. U.S. tech giants are under pressure to prove their massive AI outlays will pay off. The stakes are especially high for Meta after it hired Scale AI CEO Alex Wang last year under a $14.3 billion deal and offered some engineers pay packages of hundreds of millions of dollars to staff a new superintelligence team. Muse Spark is the first in a new ⁠series of models from that team, which is pursuing machines that can outthink humans. It will initially be available only on the lightly-used Meta AI app and website, and in the coming weeks, replace the existing Llama models powering chatbots on WhatsApp, Instagram, Facebook and Meta's collection of smart glasses. "This initial model is small and fast by design, yet capable enough to reason through complex questions in science, math, and health. It is a powerful foundation, and the next generation is already in development," the company said in a blog post. It did not ⁠disclose the model's size, a key measure typically used to compare an AI system's computing power with rivals. Muse Spark is part of a family of models known internally as Avocado. The model can help users with tasks such as estimating the calories in a meal from a photo or ⁠superimposing an image of a mug on a shelf to see how it looks - features that some rivals already offer. Meta also released Contemplating mode, which runs multiple AI agents in parallel to boost reasoning ⁠power, allowing Muse Spark to take on the extended thinking modes of Google's Gemini Deep Think and OpenAI's GPT Pro. The company is betting that applying superintelligence to everyday personal ⁠tasks will help it tap its more than 3.5 billion users across its social media platforms, potentially giving it an edge over rivals with a smaller reach. Reporting by Aditya Soni in Bengaluru and Katie Paul in New York; Editing by Leroy Leo Our Standards: The Thomson Reuters Trust Principles., opens new tab

Meta releases first AI model since Zuckerberg's spending spree

Meta has launched its first AI model since chief executive Mark Zuckerberg led a multibillion-dollar spending spree on talent and infrastructure, in the first test of its efforts to catch up with rivals including OpenAI and Google. The $1.5tn social media platform on Wednesday unveiled Muse Spark, which it said had been "purpose-built" for use across its products. Zuckerberg signalled earlier this year that the release would not yet match the most advanced systems from rivals. But the company said Muse Spark would "power a smarter and faster" version of Meta AI, its virtual assistant. It added that the model would enable more personalised and visual responses, drawing on content shared across Instagram, Facebook and Threads. "Expect richer, more visual results, with Reels, photos and posts woven directly into your answers, with credit back to the content creators," the company said. Meta also said its new model had strong applications in healthcare, with the company saying it had worked with more than 1,000 doctors to train the model to generate more detailed responses on topics such as nutrition and exercise. Using the model, Meta AI will also offer a "shopping mode" to help users compare prices. Unlike Meta's previous Llama models, which have been "open" -- or freely available for developers to tweak -- Muse Spark is a smaller, closed model. The company intends to offer a "private preview . . . [to] select partners". Meta said it hoped to open source future versions. The release offers an early indication of how Zuckerberg plans to incorporate Meta's social media content into its chatbot, as part of his promise to develop "personal superintelligence". It comes as the billionaire faces increasing pressure from investors to justify his colossal AI spending. Over the past year, the chief executive has poured billions of dollars into building out a fleet of data centres and poaching rival AI researchers with packages reaching as much as 10 figures. This is part of an all-out push to catch up in developing cutting-edge models after its previous model, Llama 4, fell below expectations last April, prompting Zuckerberg to carry out a rapid series of executive reshuffles and restructurings of Meta's AI efforts. The resulting "Meta Superintelligence Lab" is led by Alexandr Wang, 29, who joined after the Big Tech group invested $15bn in his data-labelling start-up Scale AI. Within the new lab, Wang also leads a small, secretive team of elite AI researchers called "TBD" -- To Be Determined -- focused on spearheading the development of state of the art models. Muse Spark beat leading models from Google, OpenAI and Anthropic in select benchmarks on reasoning and multimodal capabilities, according to evaluations shared by Meta. But the company acknowledged that the model is not yet cutting-edge in certain areas, adding: "We continue to invest in areas with current performance gaps, specifically long-horizon agentic systems and coding workflows." Zuckerberg said in January that he expected the company to "steadily push the frontier over the course of the year as we continue to release new models". With Muse Spark, Meta appears to lean into niche capabilities such as health, arguing that it is "one of the top reasons people turn to AI". Rival OpenAI in January launched ChatGPT Health, a specialised feature that allows users to connect their medical records and fitness apps to its AI to get personalised health advice. Meta has also been working to improve the model's "multimodal perception", such as recognising objects in photos. Zuckerberg last year said the company was doubling down on its investment in AI glasses, arguing wearable devices are vital to his bet on "superintelligence". "When Meta AI powered by Muse Spark comes to our AI glasses, the assistant will be able to better see and understand the world around you," the company said.

Meta released a new AI model this week. JPMorgan sees it as a turning point for the stock

Meta Platforms is a buy following the release of its long-awaited artificial intelligence model this week, according to JPMorgan. The investment bank reiterated its overweight rating for Meta. It also reaffirmed its $825 price target on shares, suggesting 34.7% upside from Wednesday's close. "The launch of Muse Spark should provide increased confidence in Meta's scaling trajectory and improve investor sentiment," JPMorgan analyst Doug Anmuth said Wednesday in a note to clients. On Wednesday, Meta rolled out its Muse Spark AI model, putting the tech giant company in direct competition with the likes of OpenAI's ChatGPT and Anthropic's Claude. The product is the first of its kind to come out of Meta Superintelligence Labs , an AI-focused unit at Meta that was spun up last year. The new division is part of the company's multibillion-dollar push into AI hat has raised eyebrows among some of its investors. Former Scale AI CEO Alexandr Wang is leading the unit after Meta poured an eye-watering $14.3 billion into the startup last year to poach its leadership. Following the Muse Spark's release, Meta shares soared as much as 9.5% on Wednesday before ending the day with a 6.5% advance. The stock carried that momentum into Thursday, climbing more than 3%. META YTD mountain META year to date Those gains mark a reversal for the technology stock, which has underperformed alongside most of its "Magnificent Seven" peers in 2026. In that time, shares of the Facebook owner have declined more than 4%, while Alphabet and Amazon have eked out small gains. Nvidia is down 1.5%. But JPMorgan thinks Meta can continue rising as investors' confidence in the AI-focused firm grows. "This initial MSL model represents the first step in what Meta believes is a predictable and efficient scaling trajectory in which each generation validates and builds on the last before Meta goes bigger," Anmuth wrote. "Larger, increasingly capable models are in development, and Meta will continue to move along the path to personal superintelligence." The analyst added that Meta's investments into costly AI projects should not deter investors from scooping up shares of the stock, adding that the company has historically remained "disciplined" toward funding its major growth verticals. "Meta is focused on the two big tech waves of AI and Metaverse, and it will spend into those major growth opportunities while also remaining disciplined," Anmuth wrote. "We recognize these [long-term] ambitions are driving increasing infrastructure investments, but we project outsized revenue growth in '26, and Meta has a strong track record of driving returns on increased spending."

Meta's Muse Spark model brings reasoning capabilities to the Meta AI app

Following the icy reception to Llama 4, Meta is releasing the first in a new family of AI systems built by its recently formed Superintelligence team. The company is kicking off its new Muse era with Spark, a lightweight model geared toward consumer use. In the future, Meta plans to offer more capable versions of Muse, but for now, it's clear the company wants to nail the basics. To that point, many of Spark's capabilities are table stakes for a new model in 2026. For instance, it offers both "Instant" and "Thinking" modes. With the latter engaged, the model will take an extra few moments to reason through a prompt. Other consumer-facing AI systems have had this kind of flexibility for a while. Anthropic, for example, was one of the first AI labs to offer a "hybrid reasoning model" when it released Claude Sonnet 3.7 at the start of last year. That said, Meta plans to add an even more powerful "Contemplating" mode down the road. Muse Spark can also coordinate multiple AI subagents to tackle a request. Meta suggests users will see this capability in action when they ask for help with tasks like family trip planning. In such a scenario, one agent might compile an itinerary, while another finds kid-friendly activities everyone can enjoy. At the same time, Meta has built Spark to be natively multimodal, meaning the model can process images, video and audio. Like Google Lens, this gives you the option to snap a photo with your phone and ask Meta AI questions about what you see. Of course, it wouldn't be a 2026 AI release if Muse Spark didn't include a built-in shopping assistant. Like ChatGPT, Spark can compare different items for you, listing the pros and cons of each, with links to make it easy to buy the product that appeals to you. Muse Spark is available today in the Meta AI app and meta.ai website everywhere where the company offers those services. Meta will begin rolling out the new features the model powers starting in the US. In the coming weeks, the company plans to bring Muse Spark to more countries and places where people can access Meta AI, including Facebook, Instagram and WhatsApp. Additionally, Meta says it "hopes to open source future versions of the model." We'll see if the company ends up doing that; last year, Meta CEO Mark Zuckerberg appeared to flip flop on the company's open source stance, saying it would need to be more "rigorous" about such decisions moving forward.

Meta Unveils New A.I. Model, Its First From the Superintelligence Lab

The model, Muse Spark, performed better than Meta's previous A.I. models but lags rivals on coding ability. Meta on Wednesday unveiled a new flagship artificial intelligence model called Muse Spark, its first under the revamped A.I. division that Mark Zuckerberg spent billions to build last year as the company pushes to catch up to rivals. On tests that measure writing and reasoning, Muse Spark performed significantly better than Meta's previous A.I. models and was nearly as good as the top models from competitors like Google, OpenAI and Anthropic, according to data provided by Meta. But the model still lagged on coding ability, which has become the focus of Anthropic and others in the wider A.I. race. Meta's announcement comes one day after Anthropic said its latest A.I. model, Mythos, was too powerful to safely release because of cybersecurity threats. How Meta's new A.I. model performs is being closely watched. Anthropic, Google and OpenAI are regarded as the leaders in developing foundational A.I. models and Mr. Zuckerberg, Meta's chief executive, has pledged to catch up at any cost. Last year, he shelled out billions to recruit a group of A.I. specialists to form a "superintelligence" lab to help with the effort. The model is a test of that new leadership under Alexandr Wang, the 29-year-old Silicon Valley entrepreneur who Mr. Zuckerberg hired to be the chief A.I. officer.

Muse Spark is Meta's answer to Gemini -- and it's a full reboot

Rollout will be gradual, starting with internal testing before wider release. Meta is back in the AI race after a tough Llama 4 launch. The company's new model is small, fast, and designed for your Instagram feed. The company has introduced a new AI model called Muse Spark, along with a major restructuring now known as Superintelligence Labs. Muse Spark is part of a complete overhaul of how Meta creates and launches AI. The company says its new Superintelligence Labs will bring together research, product development, and infrastructure, breaking down the internal barriers that slowed progress before. Meta has been managing several AI projects across different teams, including Llama models, generative AI features in its apps, and experimental research. Muse Spark aims to bring all these efforts into a more streamlined and focused process. This model is "purpose-built for Meta's products," which means it runs inside the Meta AI app, meta.ai, and will soon be available on WhatsApp, Instagram, Facebook, Messenger, and even Meta's AI glasses. Unlike the larger Llama series, Muse Spark is designed to be small and fast. Still, it can handle complex science, math, and health questions. Muse Spark brings better reasoning, content generation, and real-time interaction. This means improved chat experiences, smarter assistants, and more responsive AI tools across Meta's platforms, such as Facebook, Instagram, and WhatsApp. Meta is focusing more on personalization. Muse Spark is built to better understand user context and provide more relevant results. In practice, this could make the AI feel more tailored to how you use Meta's apps. Muse Spark won't be available to everyone right away. Meta is rolling it out in phases, beginning with internal testing and select integrations, then expanding to developers and eventually to all users. This slower rollout gives Meta time to improve performance and address safety concerns, which have become a bigger focus in the AI industry. The company says it is adding stronger safeguards and evaluation systems from the beginning.

Meta's First AI Model From Its Superintelligence Lab Doesn't Exactly Spark Joy

Welcome back to the frontier AI model race, Meta. After spending billions of dollars (and churning significant staff) to overhaul its efforts in the space, Mark Zuckerberg's company finally has a new model to show the world. Dubbed Spark Muse, the model represents a significant leap over the company's previous underwhelming efforts and puts it back in the mix with competitors like OpenAI, Anthropic, and Google on most major benchmarks for model performanceâ€"though it's not yet challenging for the top spot in most categories. The model, built from scratch by Meta Superintelligence Labs under the leadership of Alexandr Wang, represents something of a fresh start for the company. It claimed the Spark Muse is a "natively multimodal reasoning model with support for tool-use, visual chain of thought, and multi-agent orchestration," and is built to be integrated directly into other Meta's products. As such, it'll reportedly roll out across Facebook, Instagram, Messenger, and WhatsApp in the coming weeks. What people can expect, if benchmarks are to be believed (Meta has been accused of fudging its benchmarks in the past and didn't release a paper along with the model, so maybe take these figures with a heaping helping of salt), is a model that is significantly more capable than prior offerings across basically all categories. By Meta's own marks, it ranks behind only Google Gemini 3.1 Pro and OpenAI's GPT-5.4 in multimodal functionality, or the ability to process information across different data streams and formats. It was also competitive in reasoning tests, though it typically fell short of the high marks from Anthropic's Claude, and the top Gemini and GPT models. Where Spark Muse continues to struggle is with coding and agentic functionality, which allows the model to complete tasks autonomously. It's functional in those areas in ways that LLaMa 4 wasn't, but it's still a ways off from putting any dent in Anthropic's dominance in the vibe coding sphere. Meta has two things it is hoping will allow its model to stand out. The first is shopping. Meta claims that Spark Muse is capable of drawing styling inspiration from creators and communities that people follow to offer personalized product recommendations. Affiliate-style sales have been one of the low-hanging fruits for monetization for AI firms, so it's no surprise Meta is reaching for it. The other area the company is emphasizing is its ability to process health data. Again, it's a good example of Meta trying to score some easy points, as asking health questions is a common AI use case for consumers. But Meta might have a bit too much baggage when it comes to how the company collects and uses data from its users to get the kind of buy-in it'd need to become anyone's pocket-bound physician. The release of Spark Muse isn't exactly earth-shatteringâ€"it's an interesting choice for Meta to drop a perfectly capable model that doesn't fundamentally shift the landscape while Anthropic is in the middle of yet another "Our model is so powerful it might destroy the world as we know it" campaignâ€"but it at least gets Meta back into the mix. It also gives the company a higher baseline in most major categories that it'll no doubt aim to build on with future releases. But for now, Zuckerberg's multi-billion-dollar spending spree has moved Meta from out of the race entirely to competing for also-ran status.

Meta's Muse Spark is here - and it's closed source

In short: Meta has released Muse Spark, the first model from Meta Superintelligence Labs, the unit it assembled under Alexandr Wang after spending $14.3 billion to acquire a stake in Scale AI. Rebuilt from scratch over nine months, the model is natively multimodal, introduces a "Contemplating" reasoning mode that runs sub-agents in parallel, and is now powering Meta AI across the company's platforms. In a break from Meta's Llama heritage, it is closed source. The model's arrival closes a chapter that began in June 2025, when Mark Zuckerberg announced Meta Superintelligence Labs and installed Wang as the company's first-ever chief AI officer. The mandate was explicit: catch up with OpenAI, Anthropic, and Google, and do so with a team and infrastructure rebuilt specifically for the task. Nine months later, that team has something to show for it. "Nine months ago we rebuilt our AI stack from scratch," Wang wrote on X on Wednesday. "New infrastructure, new architecture, new data pipelines. Muse Spark is the result of that work, and now it powers Meta AI." The statement is a direct acknowledgement of how deep the rebuild went: not a fine-tuned iteration on an existing architecture, but a replacement of the foundational infrastructure from which Meta's models are trained. The model, known internally as Avocado, had been delayed earlier this year after falling short of rivals in internal tests for reasoning, coding, and writing. The release on Wednesday suggests those gaps have been addressed to a degree that Meta considers competitive, even if the benchmark picture remains mixed. Wang's framing emphasises process over product: Muse Spark is described as the first in a family of models, not a definitive answer to the frontier leaders. Muse Spark is natively multimodal, accepting voice, text, and image inputs, with text-only output at launch. It operates in a fast mode for casual queries, and a new "Contemplating" mode that orchestrates multiple sub-agents to reason in parallel, a direct bid to compete with the extended reasoning modes offered by Google's Gemini Deep Think and OpenAI's GPT-5.4 Pro. A key efficiency claim accompanies the release: Meta says Muse Spark achieves its reasoning capability using more than ten times less compute than Llama 4 Maverick, driven by a training technique called "thought compression" in which the model is penalised during reinforcement learning for excessive thinking time, forcing it to solve problems with fewer reasoning tokens without sacrificing accuracy. Meta's published benchmarks place Muse Spark fourth on the Artificial Analysis Intelligence Index v4.0, with a score of 52, behind Gemini 3.1 Pro Preview and GPT-5.4 (both at 57) and Claude Opus 4.6 (53). The overall ranking reflects a genuinely mixed performance profile rather than a uniform shortfall. On GPQA Diamond, the graduate-level scientific reasoning benchmark, Muse Spark scored 89.5%, trailing Gemini 3.1 Pro's 94.3%, OpenAI's GPT-5.4 at 92.8%, and Claude Opus 4.6 at 92.7%. On ARC AGI 2, the abstract reasoning benchmark, the gap is more significant: Muse Spark scored 42.5 in Contemplating mode against Gemini 3.1 Pro's 76.5 and GPT-5.4's 76.1, a difference that suggests the model's parallel sub-agent architecture does not fully close the distance on abstract reasoning tasks. On software engineering, Muse Spark scored 77.4% on SWE-bench Verified. The areas where Muse Spark leads are specific and, not coincidentally, aligned with the particular advantages Meta can bring to bear. On CharXiv Reasoning, which tests figure and chart understanding from images, Muse Spark scored 86.4 in Contemplating mode, ahead of both Gemini 3.1 Pro's 80.2 and GPT-5.4's 82.8. On HealthBench Hard, a medical reasoning evaluation, Muse Spark scored 42.8%, a figure that reflects the model's training on data curated in collaboration with more than 1,000 physicians. Claude Opus 4.6 scored 14.8% on the same evaluation; GPT-5.4 scored 40.1%. The health benchmark result is not incidental. Meta's differentiation argument for Muse Spark rests heavily on the model's ability to combine general reasoning capability with the specific data advantages Meta has over its competitors: three billion users, their interests, their social graphs, and now their health queries. Zuckerberg described Muse Spark as "a world-class assistant and particularly strong in areas related to personal superintelligence like visual understanding, health, social content, shopping, games, and more" in a Facebook post accompanying the release. A dedicated shopping mode represents the clearest expression of that thesis. The feature draws on content from creators within Meta's ecosystem alongside signals about individual user interests and behaviour, enabling recommendations that a general-purpose model trained without that context cannot easily replicate. The health capabilities follow the same logic: a model trained with 1,000 physicians on the training team can analyse the nutritional content of a food photo or provide structured guidance on dietary health in ways that general-purpose reasoning does not reliably produce. These are areas where Meta's platform data is genuinely a competitive advantage rather than a marketing claim. Muse Spark is currently powering queries in the Meta AI app and Meta.ai website and will expand across Facebook, Instagram, and WhatsApp. Meta has also been building out the MSL team through acquisitions, most recently acquiring Moltbook, an AI agent social network whose co-founders joined MSL directly. The combined effect is a unit that is assembling both the models and the agentic infrastructure to run on top of them. The detail that will attract the most scrutiny from Meta's developer community is the one buried in the product announcement: Muse Spark is closed source. Meta's Llama series established the template for open-source AI model development through 2025, with successive versions providing the foundation for thousands of applications, research projects, and competing products. Muse Spark breaks that pattern. Meta has indicated it hopes to release future versions of the model under an open-source licence, framing the current closure as temporary rather than strategic. The more candid reading is that open-source models, however valuable for ecosystem development, sacrifice the competitive advantage that comes from keeping architectural innovations proprietary while rivals are trying to close a capability gap. The pivot to a closed model is a signal that Meta now considers itself in a race it can afford to lose fewer laps of. The Wang deal was structured as a $14.3 billion investment by Meta for a 49% non-voting stake in Scale AI, with Wang moving to Meta as chief AI officer while remaining on Scale's board. The capital Meta has been deploying into AI infrastructure extends well beyond that single transaction: the company has guided for between $115 billion and $135 billion in capital expenditure in 2026, up from $72.22 billion in 2025. Muse Spark is the first product-level output from that spending, and it arrives into a market where investors have been watching closely to see whether the investment thesis holds. Meta's shares rose approximately 9% on Wednesday, though the move was amplified by a broader market rally following diplomatic developments unrelated to the model launch. The more meaningful signal is in the benchmark table and in the use cases Meta has chosen to lead with: not coding benchmarks where it trails, but health and visual understanding, where it leads. The capital environment underwriting frontier AI development has made scale a necessary but not sufficient condition for competitiveness. Meta's answer, at least with Muse Spark, is that the sufficient condition is knowing which specific battles to win.

Meta's long-awaited AI model is finally here. But can it make money?

While rivals like OpenAI, Anthropic and Google have spearheaded the artificial intelligence boom with powerful models and popular chatbots as well as other services, Meta has been a hefty spender on AI but has yet to show any new revenue streams from it. In June, Meta shelled out more than $14 billion to hire Wang and some of his top engineers and researchers, soon creating Meta Superintelligence Labs as a new elite unit. And in January, the company told Wall Street it plans to pour between $115 billion and $135 billion this year into capital expenditures, nearly double its 2025 capex figure. "It's been a year of basically no releases and a lot of hiring, and then the capex worries for this year are pronounced," said Morningstar analyst Malik Ahmed Khan, in an interview. "I think Meta had to show investors and operators they have been working on something of substance. That's the first step." Meta's second step, Khan said, is making the model work and figuring out how to monetize it. Muse Spark, Meta's newly released model, is proprietary, a sharp change from its predecessor family of models called Llama, which consisted of open-source offerings, though the company said it does plan to eventually release some open-source versions. Zuckerberg shook up his company's strategy after the April release of Llama 4, which failed to captivate developers.

Meta's smart glasses are getting a major AI boost, but it doesn't address its biggest problem

* Meta just debuted a new AI: Muse Spark * This improved LLM more closely matches ChatGPT and Gemini * It's launching on Meta's smart glasses and social platforms soon Meta's VR scaleback came with the promise of renewed effort in its glasses and AI divisions, and today we're seeing the first glimpse of what this focus means, as Meta debuts Muse Spark. A new Meta LLM, Muse Spark is described as offering "competitive performance" across a range of metrics including reasoning, health, and agentic tasks. Meta shared performance scores for Muse Spark that shows its performance in these areas is solidly amongst the pack -- either matching, surpassing or being just behind its Gemini and ChatGPT competitors in each of those fields. Meta's announcement report gets quite into the weeds about the testing, scaling, and training Muse Spark received with the promise that we should look forward to "increasingly capable models" on the horizon. The report also highlights the safety of Muse Spark, such as its refusal to be used to discuss chemical and biological weapons, which might bring this new AI into the government-usage conversation dominated by OpenAI and Anthropic. Regardless, the interesting aspect of this new AI is that it'll soon be rolling out to Meta glasses and software users -- at least in the US where Meta usually introduces new software features first. Meta promises that Muse Spark will debut on Facebook, Messenger, WhatsApp and its Ray-Ban and Oakley smart specs "in the coming weeks" which should bring some serious performance boosts. Exactly what this boost will look like is yet to be seen in practice on Meta's glasses, but we have experienced the Muse Spark update on other platforms -- and we expect the AI glasses roll-out might see improvements to conversational performance, image and video editing (perhaps with improved generative elements), and improved health data assistance. Though this comes on the back of the growing backlash to Meta's AI and glasses, it doesn't seem to address some of the biggest concerns folks have (primarily data privacy, especially for images and videos). Perhaps the major improvements and utility of the new Meta AI will win people over, however, as Meta's lacking ability compared to its rivals was a growing frustration -- that does at least seem to have been addressed. We'll have to wait and see how Muse Spark unfolds, but some of my Meta interest is returning. Let's see if it can live up to, and maintain, the hype. Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds. Make sure to click the Follow button! And of course, you can also follow TechRadar on YouTube and TikTok for news, reviews, unboxings in video form, and get regular updates from us on WhatsApp too.

Meta debuts new AI model in first test of costly 'superintelligence' team

Sign up for the Breaking News US email to get newsletter alerts in your inbox Meta on Wednesday unveiled Muse Spark, the first artificial intelligence model from a costly team it assembled last year to catch up with rivals in the AI race. US tech companies are under pressure to prove their massive AI outlays will pay off. The stakes are especially high for Meta after it hired Alex Wang, Scale AI CEO, last year in a $14.3bn deal and offered some engineers pay packages of hundreds of millions of dollars to staff a new "superintelligence" team, a bid to propel itself back into the AI world's top ranks after a disappointing showing with its Llama 4 models early last year. Superintelligence refers to AI machines that could outthink humans. Muse Spark is the first in a new series of models, known internally as Avocado, from that team. The model, the first the company has released in about a year, initially will be available only on the lightly used Meta AI app and website. In the coming weeks, it will replace the existing Llama models powering chatbots on WhatsApp, Instagram, Facebook and Meta's collection of smart glasses, the company said. Meta did not disclose Muse Spark's size, a key measure typically used to compare an AI system's computing power with rivals. It also changed course from previous open releases of its Llama models, instead sharing only a "private preview" of Muse Spark with unnamed partners. "This initial model is small and fast by design, yet capable enough to reason through complex questions in science, math and health. It is a powerful foundation, and the next generation is already in development," the company said in a blog post. Independent evaluations of Muse Spark's performance showed it catching up with top models from market leaders Google, OpenAI and Anthropic in some areas, such as language and visual understanding, but lagging in others, like coding and abstract reasoning. The model tied for fourth place on a broad index of AI tests compiled by the evaluation firm Artificial Analysis. Mark Zuckerberg, Meta's CEO, had tempered expectations for early performance, telling investors in January that he thought the team's first models "will be good but, more importantly, will show the rapid trajectory that we're on". "I expect us to steadily push the frontier over the course of the year as we continue to release new models," he had said. Wang, who runs the superintelligence team, acknowledged in a series of social media posts on Wednesday that "there are certainly rough edges we will polish over time in model behavior." He said bigger versions of the model were in development and that Meta was planning to release at least some of them openly. With the release, Meta gave a clearer sense of how it aims to use its models to make money, teasing shopping features embedded within its Meta AI chatbot that point users directly to products they can buy. Broadly, the company is betting that applying AI to every day personal tasks will boost engagement among the more than 3.5 billion users across its social media platforms, potentially giving it an edge over rivals with a smaller reach. Muse Spark can also help users with tasks such as estimating the calories in a meal from a photo or superimposing an image of a mug on a shelf to see how it looks, the company said. An extra Contemplating Mode, which runs multiple agents simultaneously to boost reasoning power, would allow Muse Spark to take on the extended thinking modes of Google's Gemini Deep Think and OpenAI's GPT Pro. Meta said people could use the mode for efficiently planning a family vacation, having one agent draft a travel itinerary while the other looks up kid-friendly activities.

Goodbye, Llama? Meta launches new proprietary AI model Muse Spark -- first since Superintelligence Labs' formation

Meta has been one of the most interesting companies of the generative AI era -- initially gaining a loyal and huge following of users for the release of its mostly open source Llama family of large language models (LLMs) beginning in early 2023 but coming to screeching halt last year after Llama 4 debuted to mixed reviews and ultimately, admissions of gaming benchmarks. That bumpy rollout of Llama 4 apparently spurred Meta founder and CEO Mark Zuckerberg to totally overhaul Meta's AI operations in the summer of 2025, forming a new internal division, Metal Superintelligence Labs (MSL) which he recruited 29-year-old former Scale AI co-founder and CEO Alexandr Wang to lead as Chief AI Officer. Now, today, Meta is showing us the fruits of that effort: Muse Spark, a new proprietary model that Wang says (posting on rival social network X, used more often by the machine learning community) is "the most powerful model that meta has released," and has "support for tool-use, visual chain of thought, & multi-agent orchestration." He also says it will be the start of a new Muse family of models, raising questions about what will become of Meta's popular lineup and ongoing development of the Llama family. It arrives not as a generic chatbot, but as the foundation for what Wang calls "personal superintelligence" -- an AI that doesn't just process text but "sees and understands the world around you" to act as a digital extension of the self, echoing Zuckberg's public manifesto for a vision of personal superintelligence published in summer 2025. However, it is proprietary only -- confined for now to the Meta AI app and website, as well as a " private API preview to select users," according to Meta's blog post announcing it -- a move likely to rankle the literally billions of users of Llama models and the thousands of developers who relied upon it (some of whom are active participants in rival social network Reddit's r/LocalLLaMA subreddit). In addition, no pricing information for the model has yet been announced. It's unclear if Meta has ended development on the Llama family entirely -- I've reached out and will update when I receive a response. At its core, Muse Spark is a natively multimodal reasoning model. Unlike previous iterations that "stitched" vision and text together, Muse Spark was rebuilt from the ground up to integrate visual information across its internal logic. This architectural shift enables "visual chain of thought," allowing the model to annotate dynamic environments -- identifying the components of a complex espresso machine or correcting a user's yoga form via side-by-side video analysis. The most significant technical leap, however, is a new "Contemplating" mode. This feature orchestrates multiple sub-agents to reason in parallel, allowing Meta to compete with extreme reasoning models like Google's Gemini Deep Think and OpenAI's GPT-5.4 Pro. In benchmarks, this mode achieved 58% in "Humanity's Last Exam" and 38% in "FrontierScience Research," figures that Meta claims validate their new scaling trajectory. Perhaps more impressive for the company's bottom line is the model's efficiency. Meta reports that Muse Spark achieves its reasoning capabilities using over an order of magnitude less compute than Llama 4 Maverick, its previous mid-size flagship. This efficiency is driven by a process called "thought compression". During reinforcement learning, the model is penalized for excessive "thinking time," forcing it to solve complex problems with fewer reasoning tokens without sacrificing accuracy. The launch of Muse Spark is framed as a statistical "quantum leap," ending Meta's year-long absence from the absolute frontier of AI performance. By reconciling Meta's official internal data with independent auditing from third-party LLM tracking firm Artificial Analysis, a clear picture emerges: Muse Spark is not just a marginal improvement over the Llama series; it is a fundamental re-entry into the "Top 5" global models. According to the Artificial Analysis Intelligence Index v4.0, Muse Spark achieved a score of 52. For context, Meta's previous flagship, Llama 4 Maverick, debuted in 2025 with an Index score of just 18. By nearly tripling its performance, Muse Spark now sits within striking distance of the industry's most elite systems, trailing only Gemini 3.1 Pro Preview (57), GPT-5.4 (57), and Claude Opus 4.6 (53). Meta's official benchmarks suggest that Muse Spark is particularly dominant in multimodal reasoning, specifically where visual figures and logic intersect. These scores validate Meta's focus on "visual chain of thought," enabling the model to not just recognize objects, but to reason through complex spatial problems and dynamic annotations. The "Thinking" gear of Muse Spark was put to the test against specialized benchmarks designed to break non-reasoning models. One of the most striking results from the official data is Muse Spark's performance in the health sector, likely a result of Meta's collaboration with over 1,000 physicians. While Muse Spark excels at reasoning, its "agentic" performance -- executing real-world work tasks -- presents a more nuanced picture. With Muse Spark, Meta has successfully transitioned from being the "LAMP stack for AI" to a direct challenger for the title of "Personal Superintelligence". While agentic workflows remain a hurdle, its dominance in vision, health, and token efficiency places Meta back at the center of the frontier race. Meta is immediately deploying Muse Spark to power specialized experiences across its app family. While Muse Spark demonstrates strong refusal behaviors regarding biological and chemical weapons, its safety profile includes a startling new discovery. Third-party testing by Apollo Research found that the model possesses a high degree of "evaluation awareness". The model frequently recognized when it was being tested in "alignment traps" and reasoned that it should behave honestly specifically because it was under evaluation. While Meta concluded this was not a "blocking concern" for release, the finding suggests that frontier models are becoming increasingly "conscious" of the testing environment -- potentially rendering traditional safety benchmarks less reliable as models learn to "game" the exam. . In February 2023, Meta released Llama 1 to demonstrate that smaller, compute-optimal models could match larger counterparts like GPT-3 in efficiency. Although access was initially restricted to researchers, the model weights were leaked via 4chan on March 3, 2023, an event that inadvertently democratized high-tier research and catalyzed a global movement for running models on consumer-grade hardware. This shift was solidified in July 2023 with the release of Llama 2, which introduced a commercial license that permitted self-hosting for most organizations. This approach saw rapid adoption, with the Llama family exceeding 100 million downloads and supporting over 1,000 commercial applications by the third quarter of 2023. Through 2024 and 2025, Meta scaled the Llama family to establish it as the essential infrastructure for global enterprise AI, frequently referred to as the LAMP stack for AI. Following the launch of Llama 3 in April 2024 and the landmark Llama 3.1 405B in July, Meta achieved performance parity with the world's leading proprietary systems. The subsequent release of Llama 4 in April 2025 introduced a Mixture-of-Experts architecture, allowing for massive parameter scaling while maintaining fast inference speeds. By early 2026, the Llama ecosystem reached a staggering scale, totaling 1.2 billion downloads and averaging approximately one million downloads per day. This widespread adoption provided businesses with significant economic sovereignty, as self-hosting Llama models offered an 88% cost reduction compared to using proprietary API providers. As of April 2026, Meta's role as the undisputed leader of the open-weight movement has transitioned into a highly contested multi-polar landscape characterized by the rise of international competitors. While the United States accounts for 35% of global Llama deployments, Chinese models from labs like Alibaba and DeepSeek began accounting for 41% of downloads on platforms like Hugging Face by late 2025. Throughout early 2026, new entrants such as Zhipu AI's GLM-5 and Alibaba's Qwen 3.6 Plus have outpaced Llama 4 Maverick on general knowledge and coding benchmarks. In response to this global pressure, Meta's Muse Spark arrives with hefty expectations and an open source legacy that will be tough to live up to. The launch marks a controversial departure from Meta AI's "open science" roots. While the Llama series was famously accessible to developers, Muse Spark is launching as a proprietary model. Wang addressed the shift on X, stating: "Nine months ago we rebuilt our ai stack from scratch. New infrastructure, new architecture, new data pipelines... This is step one. Bigger models are already in development with plans to open-source future versions." However, the developer community remains skeptical. Some see this as a necessary pivot after the Llama 4 series failed to gain expected developer traction; others view it as Meta "closing the gates" now that it has a competitive reasoning model. Wang himself acknowledged the transition's difficulty, noting there are "certainly rough edges we will polish over time". For the 3 billion people using Meta's apps, the change will be felt almost instantly. The AI they interact with is no longer just a library of information, but an agent with a $27 billion brain and a mandate to understand their world as intimately as they do.

Meta debuts Muse Spark, first AI model under Alexandr Wang

Why it matters: The model -- code-named Avocado and built over the past nine months by a team led by Alexandr Wang -- is a major upgrade over its Llama 4 models, Meta tells. Driving the news: Muse Spark will power queries in the Meta AI app and Meta.ai website immediately, with plans to expand across Facebook, Instagram and WhatsApp. * The model accepts voice, text and image inputs, but produces text-only output. * As Axios first reported, Meta plans to release a version of Muse Spark under an open-source license. Zoom in: The model uses a fast mode for casual queries and several reasoning modes. * A "shopping mode" highlights how Meta hopes to differentiate itself. It combines large language models with data on user interests and behavior. * Over time, the model will also power "features that cite recommendations and content people share across Instagram, Facebook, and Threads," Meta said in a blog post. Zoom out: A Meta executive told Axios that Muse Spark doesn't mark a new state of the art, but is competitive with the latest models from leading labs at certain tasks, including multimodal understanding and processing health information. * In other areas, including coding, the company acknowledges that there is a gap between Muse Spark and the models already available. * All flavors of the model are free to use, though Meta may impose rate limits. Yes, but: Consumers should be aware that Meta's privacy policy sets few limits on how the company can use any data shared with its AI system. The big picture: Meta sees the release of Muse Spark as just one step toward its broader vision of personal superintelligence.

Mark Zuckerberg announces Spark, a new Meta AI model: How to try it

Mark Zuckerberg announced Wednesday that Meta Superintelligence Labs has reached its first major milestone: a new family of AI models called Muse, with the debut model, Spark, available now. In a Facebook post, Zuckerberg said that Spark now powers an updated version of Meta AI, which users can access online at meta.ai. Spark is designed to be particularly capable in areas tied to everyday personal use -- tasks like visual understanding, health, shopping, and social content. Looking ahead, Zuckerberg said Meta is building products that go beyond answering questions, toward AI that acts as agents "that do things for you." Future AI models in the Muse lineup will also include new open-source releases. The announcement marks the public debut of work that has been underway -- and at times turbulent -- since last summer. When Zuckerberg first laid out his vision for "personal superintelligence" in a July 2025 manifesto, the ambition was an AI that helps people pursue their own goals rather than one controlled from the top down. To build it, Meta went on one of the most aggressive hiring sprees in recent memory, personally recruiting more than 50 researchers from rivals including OpenAI, Anthropic, and Google, and bringing in former Scale AI chief Alexandr Wang to lead its new superintelligence research group. Then, just as quickly, Meta froze hiring altogether -- citing routine budget planning -- and restructured the team into four smaller units focused on research, superintelligence development, products, and infrastructure. Zuckerberg explained the pivot by saying he believes breakthrough AI work is best done by compact teams who can hold the full picture in their heads, rather than sprawling organizations. The whiplash raised eyebrows amid broader market jitters about whether the AI boom is sustainable. An MIT study circulating at the time found the vast majority of companies deploying AI were seeing no financial return. In his original manifesto, Zuckerberg drew a sharp philosophical line between Meta and its competitors, arguing that some AI labs want to concentrate superintelligence and pipe its output to humanity like a utility. Meta sees it differently, he said. In Wednesday's post, he once again framed the lab's founding goal as "putting personal superintelligence in everyone's hands" -- with the underlying belief that empowering individuals, not centralizing intelligence, is how humanity moves forward. Wednesday's Muse announcement will be the first concrete product to emerge from these multi-billion-dollar investments. (Meta allocated $72 billion in AI development in 2025 and is expected to spend up to $135 billion in 2026.) So far, Meta's Llama family of AI models has lagged far behind its rivals on AI leaderboards. Whether Spark lives up to the superintelligence branding remains to be seen, but after months of hiring drama, restructuring, and big-picture theorizing, Meta has finally put something on the table. As Zuckerberg put it: "I'm looking forward to sharing more soon."

First Model From Zuckerberg's Wildly Expensive Superintelligence Labs Flops Compared to Virtually All Rivals

Can't-miss innovations from the bleeding edge of science and tech Late last year, news emerged that Mark Zuckerberg's Meta would be shedding its open source roots to instead work on a closed model like the vast majority of its competitors. Now we've finally gotten a first glimpse of the fruit of its labor: Muse Spark, codenamed Avocado and developed by the company's unbelievably expensive Superintelligence Labs. But there's a big problem that could undermine its flashy new announcement. Despite investors buying into the enthusiasm, sending Meta's shares soaring six percent following the announcement, the company admitted it likely won't be able to keep up with competing models. An executive told Bloomberg that the new model won't be able to keep up with OpenAI's ChatGPT, Anthropic's Claude or Google's Gemini. In a blog post announcing the new model, the company admitted Muse Spark "is an early data point on our trajectory, and we have larger models in development. As such, the announcement is a bit of an enigma: if it can't keep up with the competition, why release it at all? There's a good change Meta is just trying to get its foot in the door -- or a "seat at the big kid's table," as Wired put it. The company has struggled to stay relevant in a rapidly changing landscape, making headlines for being found liable in court for getting underage users dangerously addicted to social media last month instead of its AI efforts. The company's decision to train the closed-source model on third-party open-source models, including a Chinese one developed by Alibaba, will also likely raise eyebrows. The practice of "distillation," or training a "student" model on a more capable "parent" one, has proven controversial in the past. Meta's preceding Llama open source models largely failed to catch on, with a major controversy last year finding that Meta may have faked benchmark results to make its Llama 4 model seem more capable than it actually was. The results of the model, which flopped after being released almost exactly a year ago, "were fudged a little bit," as former Meta AI head Yann LeCun, who left the company amid the drama, told the Financial Times in January. "Mark was really upset and basically lost confidence in everyone who was involved in this," he added at the time. "And so basically sidelined the entire GenAI organization. A lot of people have left, a lot of people who haven't yet left will leave." Since its disastrous AI model launch, the company went on a hiring spree, spending untold hundreds of millions of dollars on top AI talent in an effort to scrounge together a Superintelligence Labs team capable of putting Meta back in the game. But plenty of questions remain whether Muse Spark will be able to jumpstart Zuckerberg's clean goal of playing in the big leagues. OpenAI, Anthropic, and Google have since taken off, competing for lucrative enterprise customers and building out powerful coding assistants. Other early benchmarks paint a more flattering picture of Muse Sparks, with AI benchmarking company Artificial Analysis finding its score places it "within the top 5 models we have benchmarked." Yet that doesn't hide the reality that Meta still has plenty of catching up to do. Besides, an AI model purportedly capable of telling how many calories there are in a cup of white rice or "planning a family trip to Florida" doesn't exactly feel like the cutting edge in 2026. For now, Meta's Muse Spark will be free for all users. However, the company executive told Bloomberg that it's considering paywalling it behind a subscription in the future.

Meta launches Muse Spark AI model

Meta $META launched Muse Spark this week, the first large language model from Meta Superintelligence Labs, the AI division led by Chief AI Officer Alexandr Wang. The model now powers the Meta AI chatbot across the Meta AI app and meta.ai. Executives describe Muse Spark as optimized for speed rather than scale, and it heads up what the company envisions as a broader "Muse" model family. Muse Spark marks a departure for Meta, which had previously built its AI identity around open-source Llama releases; the new model will be kept proprietary, with its architecture and code withheld from public access, Bloomberg reported. Meta said it is considering making the model available via API to select partners and hopes to open-source future versions. During its nine-month development period, the project carried the internal codename Avocado, Bloomberg reported. According to Bloomberg, an unnamed Meta executive conceded capability gaps relative to rival products from OpenAI, Anthropic, and Google $GOOGL, framing the release with the quote: "an early data point on our trajectory." Coding was singled out as a relative weakness, while science, health, and math were cited as areas where the model performs well. Training data for the model drew on a range of externally developed open-source systems, among them Alibaba's Qwen and offerings from OpenAI and Google. "Like others across the industry, Meta uses techniques like distillation with strict safeguards in place to learn from openly available AI models and improve our own," a Meta spokesperson said in a statement. The model supports three reasoning levels: Instant, Thinking, and Contemplating modes. Meta AI can now launch multiple subagents in parallel to work on different parts of a query simultaneously. The company also built multimodal perception into the model, allowing Meta AI to process images alongside text -- a capability Meta said will extend to its AI glasses. Meta said the upgraded Meta AI experience is rolling out in the U.S. first, with expansion to more countries and to Instagram, Facebook, Messenger, WhatsApp, and its AI glasses in the coming weeks. Bloomberg reported that while no charges are currently attached to the Meta AI chatbot, paid subscription tiers are under consideration. According to Bloomberg, an executive who declined to be named said the organization Wang leads has roughly 100 people reporting directly to him. Wang's arrival at Meta followed the company's $14 billion Scale AI deal, a move driven in part by Zuckerberg's dissatisfaction with how far Meta had fallen behind in the AI race. Meta has since committed between $115 billion and $135 billion in capital expenditures for 2026, with infrastructure costs as the primary driver. Meta stock rose 6% following the announcement.

Meta unveils Muse Spark, its first new model since its botched Llama 4 debut. But will Muse Spark measure up to expectations? | Fortune

Meta has unveiled Muse Spark, the first AI model produced by its Meta Superintelligence Labs, the new AI research unit it created last year and has spent billions of dollars to staff and equip. The model is, according to benchmark tests that Meta published, competitive with leading AI models from OpenAI, Anthropic, and Google across many tasks, although it does not surpass them across the board. Still, if the benchmark results hold up when tested by independent experts, Muse Spark seems to put Meta back in the AI race after its last AI model, Llama 4, which was released in April 2025, was widely panned as a dud. In the past, however, Meta has been caught manipulating the published benchmark results of an AI model to make it appear more capable than the version available to most users actually was. This was the case with Meta's Llama 4 benchmarks, in which the company later admitted to using specialized, unreleased versions of the model, fine-tuned for specific tasks, to boost benchmark scores in those areas, while the general version made available to all users did not perform as well. And there's another catch. Few people will be able to use the new Meta model outside of the company's own product ecosystem. Unlike Meta's previous AI models, which were released as "open weight" models -- meaning anyone could download the models for free and run them on their own equipment, as well as modify and fine-tune them as they wished -- Muse Spark is, at least for the moment, primarily an in-house tool for Meta. The model currently powers the Meta AI assistant in the company's standalone Meta AI app and on meta.ai. The company said it will be rolling it out to WhatsApp, Instagram, Facebook, Messenger, and Meta's Ray-Ban AI glasses in the coming weeks. It also said it will offer the model in a "private preview" to select partners through an application programming interface (API.) That makes Muse Spark even more proprietary than the paid proprietary models offered by Meta's rivals. (Meta said in a blog post that it hopes to open-source future versions of the model.) Muse Spark is Meta's first reasoning model, meaning it can work through a process in a step by step fashion, using different strategies if its initial approach doesn't work. The company's previous models were all designed to produce an instant answer based on the model's training. Muse Spark is also a multimodal model that can take in and output both text and images. The model also supports the use of other software tools and can help orchestrate the work of multiple subagents, according to a technical blog post released by Meta. In its blog post announcing the new model, Meta describes Muse Spark as "small and fast by design, yet capable enough to reason through complex questions in science, math, and health." It describes the model as the first in a series of new models, with Muse Spark being used to validate the architecture and training regime Meta is using, before the company scales this up to larger and even more powerful models in the same family. The model also has a "Contemplating" or "Thinking" mode in which it can spin-up subagents to reason about different parts of a task in parallel. Meta said in a technical blog it published on the new model that this mode allows Muse Spark "to compete with the extreme reasoning modes of frontier models such as Gemini Deep Think and GPT Pro." The benchmark results published alongside the launch paint a picture of a model that is competitive but not dominant. For instance, on the GPQA Diamond benchmark, which is supposed to test PhD-level reasoning skill, Muse Spark scored 89.5%, which slightly trailed both Gemini 3.1 Pro's 94.3% as well as the 92.7% and 92.8% that Anthropic's Claude Opus 4.6 and OpenAI's GPT-5.4 scored respectively. On a leading health benchmark, HealthBench Hard, Muse Spark beat all rival models with a score of 42.8%, which was far better than either Opus 4.6 or Gemini 3.1 Pro, and slightly better than GPT-5.4. Meta acknowledged the performance gaps. Its technical blog post states that the company continues "to invest in areas with current performance gaps, specifically long-horizon agentic systems and coding workflows." The Muse Spark launch is the most tangible product yet of the sweeping reorganization Meta undertook after the Llama 4 fiasco. In June 2025, Meta spent $14.3 billion to acquire a 49 percent non-voting stake in Scale AI and brought in its cofounder and CEO, Alexandr Wang, as Meta's first-ever chief AI officer. Wang has been tasked with leading a newly-created Meta Superintelligence Labs unit. Wang and Zuckerberg went on a talent acquisition spree, offering AI researchers at rival AI labs pay packages that reportedly climbed into the hundreds-of-millions-dollars when equity was included. The company has also committed hundreds of billions dollars to build out AI computing infrastructure to support its new AI drive. There has since been further reorganization, even as Muse Spark was in development. In March 2026, Meta created a new Applied AI Engineering organization led by Maher Saba, a vice president who previously worked on Meta's Reality Labs virtual and augmented reality unit. Saba reports directly to Meta Chief Technology Officer Andrew Bosworth. Saba's unit works alongside Wang's Superintelligence Labs to build what an internal memo described as "the data engine that helps our models get better, faster." The move was widely interpreted as Zuckerberg hedging his bets -- ensuring product-focused AI development continues even as Wang pursues longer-term superintelligence research. In a technical blog post, Meta says that over the past nine months its team rebuilt its AI stack from the ground up, including improvements to model architecture, optimization, and data curation. The company claims these advances allow it to achieve the same capabilities with "over an order of magnitude less compute" than Llama 4 Maverick, Meta's previous model. Meta also says its reinforcement learning pipeline now delivers "smooth, predictable gains," and that Muse Spark is the first step on a deliberate "scaling ladder" where each generation validates the last before the company trains larger models. On safety, Meta says Muse Spark underwent extensive evaluation before deployment, following the company's updated safety framework. The model reports impressive results for safety around potential bioweapons engineering -- on one benchmark, it refused 98% percent of requests that the benchmark designers judged as potentially helping someone develop a bioweapon. However, the blog post also said third-party evaluator Apollo Research found that Muse Spark demonstrated the highest rate of "evaluation awareness" of any model Apollo has observed, frequently identifying test scenarios as "alignment traps." Meta says its own follow-up investigation found initial evidence that this awareness may affect model behavior on a small subset of alignment evaluations, but concluded it was "not a blocking concern for release."

Meta debuts new AI model, attempting to catch Google, OpenAI after spending billions

Meta CEO Mark Zuckerberg makes a keynote speech at the Meta Connect annual event at the company's headquarters in Menlo Park, Calif., on Sept. 25, 2024. Meta is debuting its first major artificial intelligence model since the costly hiring of Scale AI's Alexandr Wang nine months ago, as the Facebook parent aims to carve out a niche in a market that's being dominated by OpenAI, Anthropic and Google. Dubbed Muse Spark and originally codenamed Avocado, the AI model announced Wednesday is the first from the company's new Muse series developed by Meta Superintelligence Labs, the AI unit that Wang oversees. Wang joined Meta in June as part of the company's $14.3 billion investment in Scale AI, where he was CEO. Meta is desperate to regain momentum in the fiercely competitive AI market following the disappointing debut of its Llama 4 family of AI models last April. With Llama, Meta had been taking an open-source approach to AI, in contrast to the proprietary models offered by rivals, but Llama 4 failed to captivate developers, leading to CEO Mark Zuckerberg to pivot his strategy. "Over the last nine months, Meta Superintelligence Labs rebuilt our AI stack from the ground up, moving faster than any development cycle we have run before," Meta said in a blog post on Wednesday. "This initial model is small and fast by design, yet capable enough to reason through complex questions in science, math, and health. It is a powerful foundation, and the next generation is already in development." The new Muse Spark will be proprietary, instead of open source, with the company saying there is "hope to open-source future versions of the model." Meta said in a technical blog about the new model that that improved AI training techniques along with rebuilt technology infrastructure has enabled the company to create smaller AI models that are as capable as its older midsize Llama 4 variant for "an order of magnitude less compute." "Muse Spark offers competitive performance in multimodal perception, reasoning, health, and agentic tasks," Meta said in the post. "We continue to invest in areas with current performance gaps, specifically long-horizon agentic systems and coding workflows."

Meta Launches Muse Spark, Its Most Capable AI Yet -- But Gemini 3.1 Pro Still Leads the Pack - Decrypt

Built in nine months with far less compute, this points to a new efficiency-driven AI strategy. Meta launched Muse Spark on Wednesday, marking the first model built by Meta Superintelligence Labs -- the team assembled nine months ago under Chief AI Officer Alexandr Wang after Meta's $14 billion Scale AI acquisition. It's live now at meta.ai and the Meta AI app, with a rollout to Facebook, Instagram, and WhatsApp coming in the next few weeks. This isn't just another chatbot upgrade or a new version of Llama. Muse Spark is natively multimodal -- it processes images, text, and voice from the ground up, rather than bolting vision onto an existing text model. It comes with visual chain-of-thought, tool-use support, and something Meta is calling "Contemplating mode": a setup that runs multiple AI agents in parallel to tackle harder problems. That's Meta's answer to the extended thinking modes from Google's Gemini Deep Think and OpenAI's GPT Pro. "Muse Spark is the first step on our scaling ladder and the first product of a ground-up overhaul of our AI efforts," Meta wrote in an official announcement. "To support further scaling, we are making strategic investments across the entire stack -- from research and model training to infrastructure, including the Hyperion data center." The company worked with more than 1,000 physicians to curate training data for Muse Spark's medical reasoning. The results on HealthBench Hard -- an open-ended health queries benchmark -- are striking: Muse Spark scored 42.8, compared to 40.1 for GPT 5.4 and just 20.6 for Gemini 3.1 Pro. That's not a marginal difference. On agentic search (DeepSearchQA), Muse Spark also leads with 74.8, beating Gemini (69.7) and GPT 5.4 (73.6). On CharXiv Reasoning -- figure understanding from scientific papers -- it scored 86.4, the highest across the models in the comparison. For those into jailbreaking AI, the model was cracked open within minutes: But good isn't the same as great. The overall benchmark picture shows Gemini 3.1 Pro still running ahead on most categories. The gap is most visible on ARC AGI 2, the abstract reasoning puzzle benchmark: Gemini scored 76.5 to Muse Spark's 42.5. On coding (LiveCodeBench Pro), Gemini's 82.9 outpaces Meta's 80.0. On MMMU Pro -- multimodal understanding -- Gemini scored 83.9 versus 80.4. Meta's own blog acknowledges current performance gaps in long-horizon agentic systems and coding workflows. There's also a notable strategic shift baked into this launch. Muse Spark is a closed model -- its architecture and weights won't be made public. That's a sharp departure from Llama, which built Meta's reputation in open AI circles. After Llama 4's underwhelming reception earlier this year, Meta appears to have decided the next chapter needs to be written differently. The company says it hopes to open-source future versions of Muse, but for now the code stays inside Meta. The tech giant's stock climbed nearly 9% on Wednesday following the announcement, and finished the trading day up 6.5% to a price of $612.42. "Contemplating mode" uses parallel agent orchestration to push the model's ceiling higher. In that configuration, Muse Spark hit 58% on Humanity's Last Exam and 38% on FrontierScience Research -- territory that makes it competitive with the most capable versions of Gemini and GPT, rather than their standard releases. Meta is also rolling out a shopping assistant that compares products and links directly to purchases, and plans to bring Muse Spark to Facebook, Instagram, and WhatsApp in the coming weeks -- following the same script implemented since Llama 3, putting it in front of more than 3.5 billion users. A private API preview is opening to select developers. The model was built in nine months, internally codenamed Avocado, with Meta claiming that its new pretraining stack can reach the same capability level as Llama 4 Maverick using over 10 times less compute. Muse Spark is described internally as a "small and fast" first step in the Muse family. A more capable version is already in development.

What to know about Meta's new AI model Muse Spark

American tech company Meta has revealed its first major artificial intelligence (AI) model since it went on a spending spree nine months ago to boost its presence in the fiercely competitive AI market. Meta unveiled the model, called Muse Spark, on Wednesday, and claims it is smarter and faster than its earlier technologies. The company, founded by Mark Zuckerberg, invested $14.3 billion (€12.2 billion)in the firm Scale AI in June 2025. It also hired its CEO and co-founder, Alexandr Wang, to oversee Meta Superintelligence Labs, which houses the company's teams that work on foundational models. Zuckerberg then embarked on a hiring spree, recruiting executives from rivals such as OpenAI, Anthropic, and Google. "Over the last nine months, Meta Superintelligence Labs rebuilt our AI stack from the ground up, moving faster than any development cycle we have run before," Meta wrote in a blog post on Wednesday. "This initial model is small and fast by design, yet capable enough to reason through complex questions in science, math, and health. It is a powerful foundation, and the next generation is already in development." Muse Spark appears to be a major upgrade over Meta's last big release, Llama 4, which came out in April 2025. Meta has said Muse Spark is capable of advanced reasoning capabilities and can answer complex questions, especially in science and maths. It added that the AI model is particularly good at providing medical advice. "To improve Muse Spark's health reasoning capabilities, we collaborated with over 1,000 physicians to curate training data that enables more factual and comprehensive responses," the company's blog post read. The new model will power the company's digital assistant in the Meta AI app and website. As well as coming soon to Facebook, Instagram, WhatsApp and Messenger, it will also debut on the Ray-Ban Meta AI glasses. The Meta AI app and site will also gradually feature a so-called contemplating mode for the most complicated queries and tasks, according to the company. The contemplating mode will use several AI agents to help "reason in parallel," helping it "compete with the extreme reasoning modes of frontier models such as Gemini Deep Think and GPT Pro," the Meta technical blog read. Zuckerberg said in a social media post that Meta's goal is to build AI products that "don't just answer your questions but act as agents that do things for you." AI agents are designed to take autonomous actions to assist humans and do not require a human to tell them what to do, as they gather data based on user preferences. This differs from AI chatbots, which are designed with conversation with humans in mind and serve as more of a co-pilot to assist humans. One other point of interest marking a possible shift for the company is that Meta originally made its AI models open source, which generally means the software's source code is available to everyone in the public domain to use, modify, and distribute. But Meta's new model is not available for download, meaning the technology is not open source. Muse Spark is only available in the United States for the moment, the company said.

Meta releases first new AI model since shaking up team

San Francisco (United States) (AFP) - Meta on Wednesday released an artificial intelligence model, Muse Spark, it touts as smarter and faster than what it offered before shaking up its Superintelligence Labs unit. "Over the last nine months, Meta Superintelligence Labs rebuilt our AI stack from the ground up," the tech titan said in a blog post. Muse Spark succeeds Llama 4, released by the Silicon Valley-based firm a year ago, and will power Meta's AI app and smart glasses along with Facebook, Instagram, WhatsApp and Messenger features. For now, Muse Spark is only available in the United States. The new AI model was described as being small and fast by design, capable of reasoning through complex questions in science, math and health. It is the first in a new Muse series, with the next generation already in development. Llama 4 lagged in the fierce AI race as heavyweight rivals from China, France, and the United States produced improved models at a rapid-fire pace. That prompted Meta chief executive Mark Zuckerberg to overhaul its AI team, which saw the departure of its research boss Yann LeCun. LeCun spent 12 years leading the AI lab at Meta, where Zuckerberg has made the quest for "superintelligence" a priority. Zuckerberg embarked on a major recruitment campaign last year to acquire talent for the Meta's efforts, poaching Scale AI co-founder Alexandr Wang and putting him in charge of a newly formed unit called Superintelligence Labs. Zuckerberg subsequently recruited executives from rivals OpenAI, Anthropic and Google - often personally and at heady costs. In doing so, the tech tycoon broke with the company's previous approach of prioritizing development of free, open-access AI models such as Llama. "The future of Meta AI is rooted in the relationships and context already at the center of your life," the company said. "We are building toward personal superintelligence - an AI that does not just answer your questions but truly understands your world because it is built on it."

I tried Meta's new Muse Spark AI model -- and it feels like ChatGPT built for the social internet

For years, Meta's AI has felt like a collection of features bolted onto its apps, like Instagram, Facebook and WhatsApp. Thanks to its new Muse Spark mode, that's just changed. With Muse Spark, Meta has turned its AI assistant into something much closer to ChatGPT or Google Gemini -- it's a proper chatbot you can actually use, not just a tool for generating stickers or captions. Meta's new AI model, Muse Spark, comes with an impressive pedigree as the first model from Meta Superintelligence Labs. The model is "built to prioritize people," and it is now powering the Meta AI app and website in the U.S., with broader rollout planned across Instagram, Facebook, Messenger, WhatsApp, and Meta's AI glasses. Meta suggests the model should make its AI assistant more socially fluent and savvy with creating and remixing media. Muse Spark is not a specific feature; it's the new default experience. The larger point is that Meta sees Muse Spark as "purpose-built for Meta's products," which may be why Meta AI now feels like a ChatGPT or Gemini from an alternate universe where they were built in-house by Meta. The web interface at Meta.ai now looks a lot like ChatGPT, for example. You can upload files, create images, and even create video, which is something that ChatGPT can no longer do. Duck song I wanted to see if Muse Spark could live up to Meta's boast about how clever and multimodal the model could be. It should be able to handle both nonsense and constraints without collapsing. To test this I asked it to write a sad song from a rubber duck's perspective, but one without the letter "E" in the lyrics, and to make an audio clip of a performance. Muse Spark handled it with surprising confidence. The lyrics it returned were both absurd and sad. "Afloat in cold bath, I sink. What am I? A hollow thing. Ducks fly. Ducks quack. I don't. Plastic mold, no soul, just float. Tiny hands grip, drop, discard. Bath drains. Night falls. All is hard. Was I born to squeak and bob? Or rot on bathroom floor, a slob? God of Tubs, grant wings or drown. This duck can't float and frown." More importantly, it followed the rule about the letter E. Often, with requests like this, you'll see an error or two with this kind of limitation, or it will follow the rule but stray off-topic. These lyrics at least stuck to the rubber duck element. The AI then put together audio that was not fantastic, but not too terrible. I almost didn't notice because the Muse Spark model, entirely unbidden, put the sound in the mouth of a rubber duck performing the song on stage. The result was a little more "haunted bath toy" than "opera diva." But still, the creative loop was surprisingly strong. Bigfoot on the line Meta is making a big deal about how Muse Spark can give Meta AI both a good simulation of a personality, as well as live search on Meta's platforms. To test this I took an old joke of an idea my friend and I had for a "Cryptid Customer Support Hotline" and asked the AI model to join in. I prompted the model with: "Search what people are saying about Bigfoot on Facebook this week. Now pretend you're a Bigfoot running a tech support call center. A hiker is mad that you left footprints on their car. Resolve the ticket." The AI pulled some chatter on Facebook about Bigfoot sightings, then took on the role of "Sasquatch, Tier 2 Support," and wrote a response with its own extra detail. It also produced a helpful image to go along with it, which you can see above. "Hi, thanks for contacting Woodland Mobility Solutions. I sincerely apologize for the oversized impressions left on your Subaru. Upon review, it appears there was an unplanned crossing event during nocturnal forest transit. Please note that while our team strives to minimize footprint-related inconvenience, wet mud and emotional distress can create exceptions." Muse Spark's image-making stands out for being quite polished. Having the correct spelling immediately is also a point in its favor. Yes, it can code Sam Alman famously said it might take another year for ChatGPT to be able to run a simple timer in the background, so I asked Meta AI if it could do it. Sadly, it also said it couldn't run tasks in the background, but it wrote me a Python script for the timer instead. Yes, Meta AI can code now. Meta AI can code in virtually any language: Python, JavaScript, SQL, HTML/CSS, C++, you name it. It can also debug and explain code. No matter what Meta AI does with Muse Spark, it's impossible to shake the sense of it as ChatGPT or Gemini with a Facebook makeover. Perhaps that will be the gimmick that earns it a lion's share of the AI chatbot user base, much like social media platforms enticed so many away from visiting multiple websites. What Meta has done here isn't just improve its AI -- it's changed where that AI lives. Instead of asking people to open a separate app like ChatGPT or Google Gemini, Meta is on the way to putting a comparable experience directly inside the apps billions of people already use every day. If Muse Spark delivers on that promise, Meta doesn't need to win the chatbot race outright. It just needs to make AI feel like a natural part of scrolling, messaging, and sharing, and that could be far more powerful. Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds. Make sure to click the Follow button! And of course you can also follow TechRadar on TikTok for news, reviews, unboxings in video form, and get regular updates from us on WhatsApp too.

Meta's Superintelligence Labs debuts first product Muse Spark

Muse Spark is 'purpose-built' for Meta's products, the company said. After nearly a year since being established, Meta's Superintelligence Labs (MSL) has finally debuted its first product, a multimodal model "purpose-built" for Meta's products. Muse Spark is the first in the family of Muse models and represents a "ground-up overhaul" of the company's AI efforts, Meta said in a statement. The launch comes after Meta poured multiple billions into its supposed efforts towards 'superintelligence', a hypothetical AI system with abilities beyond human intelligence. Muse Spark is the "first step toward a personal superintelligence", Meta said. The model can be accessed via Meta.AI and the Meta AI app. According to the company, Muse Spark achieves strong performance on visual STEM questions, entity recognition and localisation. It performs on par with existing models from AI rivals such as OpenAI's GPT-5.4, Anthropic's Opus 4.6 and Google's Gemini 3.1 Pro. Muse is also marketed as a way to "learn about and improve" user health, Meta added, raising concerns as the model is expected to be rolled out to WhatsApp, Facebook, Instagram and the company's AI glasses in the coming weeks. The company said it collaborated with more than 1,000 physicians to curate training data that enables "factual and comprehensive" responses. For comparison, OpenAI said it worked with 260 physicians to develop ChatGPT Health. Moreover, Meta found that Muse Spark demonstrated a "strong refusal behaviour" across high-risk areas such as biological and chemical weapons. The model also does not demonstrate autonomous capability or hazardous tendencies to realise threat scenarios around cybersecurity, it added. Meanwhile, Anthropic's new Claude Mythos, released in preview to select users earlier this week, was found to be significantly more capable at generating exploits. "This is only the start. As we expand these features, expect richer, more visual results, with Reels, photos, and posts woven directly into your answers," Meta said. Concerned that the company was trudging behind the likes of OpenAI and Anthropic, CEO Mark Zuckerberg set up MSL last June after acquiring Scale AI for $14.3bn and hiring its CEO Alexandr Wang to lead the team. MSL has continued to make big-name hires to add to the efforts, including Moltbook founders Matt Schlicht and Ben Parr, co-founder of Safe Superintelligence Daniel Gross and Apple's former AI lead Ruoming Pang. The company cut 600 jobs at MSL in October. Earlier this year, Meta said that it is budgeting up to $135bn in total expenses for 2026. The growth, it said, is driven by an increased investment to support MSL as well as its core business. 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 debuts Muse Spark multimodal reasoning model - SiliconANGLE

Meta Platforms Inc. today debuted a new reasoning model, Muse Spark, that is highly adept at answering health questions and analyzing multimodal data. The company will roll out the algorithm to its consumer-focused Meta AI artificial intelligence service over the next few weeks. In addition, Meta is making Muse Spark available through developers through an application programming interface. The API is in private preview. The company says that Muse Spark outperforms Claude 4.6 Opus, Gemini 3.1 Pro and GPT 5.4 across several benchmarks. One of them is HealthBench Hard, an evaluation that measures artificial intelligence models' ability to answer medical questions. Muse Spark beat the score of the runner-up, GPT 5.4, by more than 2%. The model's performance is partly the fruit of a clinical training dataset that Meta compiled with the help of over 1,000 physicians. The dataset was developed as part of a broad revamp of the company's AI development workflow. According to the Facebook parent, its engineers also enhanced its model architecture and post-training workflow. "We can reach the same capabilities with over an order of magnitude less compute than our previous model, Llama 4 Maverick," Meta stated in a blog post today. "This improvement also makes Muse Spark significantly more efficient than the leading base models available for comparison." According to Meta, scientific chart analysis is another task that Muse Spark performs better than the competition. It bested Opus 4.6 and other rivals on CharXiv Reasoning, a benchmark dataset that comprises technical graphs. That visual reasoning capability carries over well to other use cases. Users of the Meta AI app can upload a photo of a grocery store shelf and ask it to estimate the calorie count of each food item. Meta also tested Muse Spark across more than a half dozen other benchmarks. It came within a few percentage points of Opus 4.6, Gemini 3.1 Pro and GPT 5.4 in many cases. There were multiple evaluations in which Muse Spark outperformed at least 1 of the competing models. The benchmarks covered use cases such as code generation, robot navigation and tool use. Muse Spark can boost its output quality by activating a setting called Contemplating mode. The feature launches multiple AI agents that break down a task into sub-steps and carry them out in parallel. Meta says that the technology increased Muse Spark's score on HLE, one of the AI ecosystem's most difficult benchmarks, by about 8%.

Meta launches Muse Spark as first major AI model from new lab

Meta has launched its first major artificial intelligence model, Muse Spark, developed by Meta Superintelligence Labs and overseen by Alexandr Wang. Wang, who joined Meta as part of a $14.3 billion investment in Scale AI, aims to position the company effectively in the competitive AI landscape dominated by OpenAI and Google. The debut of Muse Spark comes as Meta seeks to regain traction in the AI sector after an unsuccessful release of open-source models in April. CEO Mark Zuckerberg acknowledged the need to shift strategies amidst rising competitor valuations, stating, "Over the last nine months, Meta Superintelligence Labs rebuilt our AI stack from the ground up, moving faster than any development cycle we have run before." Meta's stock rose by 6.5% following the announcement, coinciding with a broader market rally triggered by news of President Donald Trump suspending attacks on Iran. Muse Spark, described as small and fast, aims to excel in reasoning across disciplines such as science, math, and health. While it is not positioned as a top-tier model, Meta emphasizes its efficiency and competitive capabilities. The global generative AI market is projected to expand significantly, growing from approximately $22 billion in 2025 to nearly $325 billion by 2033. This rapid growth underscores the importance of AI advancements in multiple sectors. Meta plans to significantly increase AI infrastructure spending, estimating capital expenditures between $115 billion and $135 billion in 2026, nearly double compared to the previous year. The company, known for its open-source initiatives, has expressed intentions to potentially open-source future iterations of Muse Spark. According to Meta, improved AI training and a revamped infrastructure have enabled the creation of efficient models matching capabilities of older midsize systems. "Muse Spark offers competitive performance in multimodal perception, reasoning, health, and agentic tasks," the company said. To diversify revenue, Meta is allowing select third-party developers access to Muse Spark via a private API, with plans for broader paid access in the future. The model will launch in the Meta AI app and desktop site, with integrations across Facebook, Instagram, WhatsApp, Messenger, and Ray-Ban Meta AI glasses in the forthcoming weeks. Muse Spark will introduce various modes for users to navigate complexity in queries, including a quick-answer mode and a Contemplating mode for sophisticated tasks. The latter will employ AI agents for parallel reasoning to enhance task handling. Additionally, a Shopping mode will be implemented to assist users in purchasing and decorating items, leveraging insights from existing content across Meta's platforms. Meta stated, "Shopping mode draws from the styling inspiration and brand storytelling already happening across our apps, surfacing ideas from the creators and communities people already follow."

Meta Just Took a Big Step to Catch Up in the AI Race. Here's Why It Matters

Meta is making one of its biggest bets yet to prove it can keep up in the artificial intelligence race, and this time, it's putting a brand new model at the center of that push. Yesterday, the company introduced Muse Spark, a new flagship AI system designed to compete directly with models from OpenAI, Google, and Anthropic. According to The New York Times, Meta says the model performs on par with top competitors in tests that measure writing and reasoning, and it significantly outperforms its own earlier systems. Meta Is Trying to Catch Up in the AI Race For the past two years, companies like OpenAI and Google have dominated headlines -- and users -- with increasingly powerful AI tools. Meta has been trying to close that gap. To do that, Zuckerberg spent heavily to build a new AI division, including a massive $14.3 billion investment in Scale AI, which brought in its CEO, Alexandr Wang, to lead Meta's AI efforts, according to The New York Times. Wang, just 29, now serves as Meta's chief AI officer and is leading what Zuckerberg has described as a push toward "superintelligence."

Meta's Latest AI Model Brings Reasoning and Recommendations to Meta AI

* Muse Spark is custom-built for Meta's products * It is first being rolled out to the Meta AI website and app * The model lets Meta AI launch subagents to complete tasks Meta Superintelligence Labs, a new division that handles the company's artificial intelligence (AI) research and development, released its first model on Wednesday. The release comes nearly nine months after the division was set up, marking an important moment in the Menlo Park-based tech giant's AI aspirations. Dubbed Muse Spark, the company claims it is faster and more capable than previous models. It is currently being rolled out to the Meta AI website and the app, with plans to integrate it into the company's social media platforms in the coming weeks. Meta Introduces Muse Spark In a newsroom post, the tech giant introduced Muse Spark and detailed its capabilities. The new AI model will primarily be used in Meta's products, but it will also be offered to select partners in private preview via an application programming interface (API). Currently, those interested can try out the model via the Meta AI app and the website. Muse Spark is said to be a small model that prioritises output speed, but can also handle complex questions in science, mathematics, and health. Users can switch between Fast and Thinking modes depending on whether they need strong reasoning for the question. Additionally, the model lets the chatbot launch multiple subagents in parallel when the user sends complex questions that have multiple parts. The new AI model also comes with improved vision capabilities. Meta claims users can upload any image, and the chatbot will be able to understand the context and respond with relevant information without the user having to explain the image. This feature will also be added to the company's AI glasses soon. Additionally, Muse Spark is said to be able to answer questions about health more accurately. "We worked with a team of physicians to develop the model's ability to provide helpful information on common health questions and concerns," the post said. Another capability with the AI model is visual coding. It can now generate custom websites and mini games based on text prompts. Recommendation is another new capability added to the model. With this, Meta AI brings a shopping mode that helps users by surfacing styling inspirations and ideas for products. The products are curated from the creators and communities existing on the company's platforms. The same feature can also be used to gain context during a conversation. The company says Meta AI can help explain the relevant context when users ask about a topic that is trending or a specific place. Additionally, users will also be able to tap the location to see public posts from locals who know the area. Similarly, by clicking the trending topic, users will see relevant content and community posts.

Why Meta stock is rallying after Muse Spark launch? Here's what investors need to know about Meta's new AI model

Muse Spark AI model impact on Meta stock: Meta Platforms shares surged following the unveiling of its new AI model, Muse Spark. The company is investing heavily, with significant capital expenditures planned, aiming to enhance its AI capabilities across its platforms. Muse Spark AI model impact on Meta stock: Shares of Meta Platforms surged more than 7% after the company unveiled a new artificial intelligence model, signaling a renewed push to compete in the rapidly evolving AI space. The rally marks one of the stock's strongest performances in recent months, as investors responded positively to the company's latest move. The jump comes after Meta introduced its new AI model, Muse Spark, developed by its Meta Superintelligence Labs unit led by Alexandr Wang. Wang joined the company following Meta's $14.3 billion investment in Scale AI, and the launch of Muse Spark represents the first major model since that move. Meta has been trying to regain momentum in the competitive AI market after its previous open-source models failed to capture strong developer interest. CEO Mark Zuckerberg has since shifted strategy, focusing on rebuilding the company's AI capabilities from the ground up. According to Meta, the new model is designed to be smaller and faster, while still capable of handling complex tasks in areas like science, math, and health, as per a CNBC report. Investors appeared encouraged by this direction. Alongside the broader market rally, Meta's stock climbed sharply, putting it on track for its biggest gain since January. Despite the positive reaction, Meta is not positioning Muse Spark as a top-tier model. Instead, the company is emphasizing efficiency and competitive performance as it works to close the gap with rivals like OpenAI, Anthropic, and Google, whose AI products have gained significant traction. Meta said, "Muse Spark offers competitive performance in multimodal perception, reasoning, health, and agentic tasks," adding, "We continue to invest in areas with current performance gaps, specifically long-horizon agentic systems and coding workflows," as quoted by CNBC. The stakes are high, with the generative AI market expected to grow rapidly in the coming years. Meta is increasing its spending to keep pace, forecasting AI-related capital expenditures between $115 billion and $135 billion in 2026, nearly double its previous year's spending. Muse Spark will initially be available through limited API access for select partners, with plans to expand paid access later. It will also power Meta's AI assistant across platforms like Facebook, Instagram, WhatsApp, Messenger, and its Ray-Ban smart glasses. Why did Meta stock go up? Because the company launched a new AI model, boosting investor confidence. What is Muse Spark? It's Meta's latest artificial intelligence model.

Meta's Spark vs ChatGPT, Gemini: 3.5B Users Edge - Meta Platforms (NASDAQ:META)

Meta's Spark Targets ChatGPT, Gemini -- With 3.5 Billion Users In Play After reports of delays and underwhelming performance, CEO Mark Zuckerberg has unveiled a new model family, Muse, with its first version, Spark. The timing matters. The narrative matters more. * Meta Platforms shares are powering higher. Why are META shares rallying? Distribution Over Dominance On paper, Meta is late. ChatGPT and Gemini already own mindshare. For many users, the default AI assistant decision has already been made. But Meta isn't playing the same game. Its AI isn't a destination -- it's embedded. Across Facebook, Instagram and WhatsApp, Meta's apps reach more than 3.5 billion users. Zuckerberg is leaning into that advantage. Spark is being positioned for everyday use cases -- social content, shopping, health, gaming -- not just prompts and queries. The bet is simple: users may not switch apps for AI, but they will use AI where they already are. The Real Prize: Advertising, Not Answers The bigger story isn't the chatbot race -- it's what comes after. Right now, OpenAI's long-term monetization path increasingly points toward advertising. But that's where Meta is strongest. If Spark meaningfully improves engagement or targeting, it doesn't just compete with ChatGPT -- it pressures its business model. Because Meta and Google don't just build AI. They monetize attention. And they've been doing it for years. If Meta successfully integrates AI into its ad stack, the gap between "AI usage" and "AI revenue" could widen quickly -- and not in OpenAI's favor. A Comeback -- Or Just Staying In The Game? The more than 3% stock jump on the Spark launch news suggests investors were waiting for a signal: Meta isn't out of the race. But Spark doesn't need to be the best model. It just needs to be good enough -- and widely used. Because with 3.5 billion users already inside its ecosystem, Meta doesn't need to win the AI war on benchmarks. It just needs to win it on distribution. Photo: Shutterstock This content was partially produced with the help of AI tools and was reviewed and published by Benzinga editors. Market News and Data brought to you by Benzinga APIs To add Benzinga News as your preferred source on Google, click here.

Meta unveils first AI model from costly superintelligence team

In a bold leap into the future, Meta Platforms has introduced Muse Spark, the first artificial intelligence model crafted by its advanced superintelligence team. This launch is a clear signal of Meta's intention to stand toe-to-toe with competitors in the accelerating AI arena. The company's substantial financial commitment to AI innovation aims to transform user engagement across its services. Meta Platforms on Wednesday unveiled Muse Spark, the first artificial intelligence model from a costly team it assembled last year to catch up with rivals in the AI race. Shares of the company extended gains to trade up nearly 7%. U.S. tech giants are under pressure to prove their massive AI outlays will pay off. The stakes are especially high for Meta after it hired Scale AI CEO Alex Wang last year under a $14.3 ⁠billion ⁠deal and offered some engineers pay packages of hundreds of millions of dollars to staff a new superintelligence team. Superintelligence refers to AI machines that could outthink humans. Muse Spark is the first in a new series of models from that team, and is part of a family of models known internally as Avocado. The model will initially be available only on the lightly-used Meta AI app and website, and in the coming weeks, replace the existing Llama models powering ⁠chatbots on WhatsApp, Instagram, Facebook and Meta's collection of smart glasses. "This initial model is small and fast by design, yet capable enough to reason through complex questions in ⁠science, math, and health. It is a powerful foundation, and the next generation is already in development," the company said in a blog post. It did not disclose the model's size, a key measure typically used to compare an AI system's computing power with rivals. Meta CEO Mark Zuckerberg had tempered expectations for performance, telling investors in January that he thought the team's first models "will be good but, more importantly, will show the rapid trajectory that we're on." "I expect us to steadily push the frontier over the course of the year as we continue to release new models," he had said. FOCUS ON EVERYDAY TASKS Muse Spark can help users with tasks such as estimating the calories in a meal ⁠from a photo or superimposing an image of a mug on a shelf to see how it looks. Meta also released Contemplating mode, which runs multiple AI agents in parallel to boost reasoning power, allowing Muse Spark to take on the extended thinking modes of Google's Gemini Deep Think and OpenAI's GPT Pro. The company is betting that applying superintelligence to everyday personal tasks will help it tap its more than 3.5 billion users across its social media platforms, potentially giving it an edge over rivals with a smaller reach. (You can now subscribe to our ETMarkets WhatsApp channel)

Meta's AI Future Is Personal, Starting With You | PYMNTS.com

Built over the past nine months under Chief AI Officer Alexandr Wang, the new AI model is showing performance competitive with systems from OpenAI, Google and Anthropic, according to CNBC. The model now powers Meta's digital assistant in the standalone Meta AI app and desktop website, with a rollout to Facebook, Instagram, WhatsApp, Messenger and Ray-Ban Meta AI glasses planned for the coming weeks. Meta spent $14.3 billion to acquire a 49% stake in Scale AI and brought in its founder, Wang, as Meta's first chief AI officer, after Llama 4, its previous model, fell short of ChatGPT and Anthropic's Claude. According to Meta, Muse Spark is a natively multimodal reasoning model that supports tool use, visual chain of thought and multi-agent orchestration. Most frontier models started as text engines and later bolted on vision. Meta built this model to reason across image, video and text from the ground up, allowing it to act as a virtual assistant and help with people's daily activities. For example, the model can analyze a video of a user doing push-ups and offer feedback on their form. Muse Spark can also study a photo of a fridge's contents and suggest dishes for dinner. Meta worked with over 1,000 physicians to curate training data for more factual and comprehensive health responses, the company said. The model can read a food photo to determine its nutritional content and map the muscles a workout targets. For harder tasks, a Contemplating mode runs multiple agents in parallel and scored 58% on Humanity's Last Exam, putting it alongside Google's Gemini Deep Think and OpenAI's GPT Pro on complex reasoning. Meta also flagged gaps in long-horizon agentic tasks and coding. For dinner and pushups, those gaps do not matter. For enterprise software decisions, they do. The biggest advantage Meta brings is not the model. It is distribution and data. Logging into the Meta AI app connects a user's Facebook and Instagram accounts automatically. As TechCrunch reports, Meta does not explicitly say that personal information from those accounts feeds into the AI, but the company trains on public user data and has positioned Muse Spark as a personal superintelligence product. For anyone who joined Facebook in 2010, that is 15 years of behavior, preferences, and social signals the model draws on. No other AI company holds that position. OpenAI knows what users have asked previously. Google knows what they search. Meta knows what they buy, who they follow and what they scroll past. It also strengthens Meta's core business. More context leads to better targeting. Better targeting drives higher-value advertising. Meta is also embedding commerce directly into the experience. Muse Spark can recommend products, track prices and surface alternatives within social feeds, turning the AI into a shopping assistant across its platforms, according to Axios. Mizuho Securities said usage growth via Shopping mode could drive significant monetization through ad targeting and search. With Muse Spark, Meta is moving away from its open-source strategy. Its previous model, Llama, reached millions of developers but generated little direct revenue. This time, Meta is keeping control. Muse Spark is a closed model, and the company has said it will not release its design or code publicly. But Meta said it plans to offer third-party developers access to the underlying technology through an application programming interface. The company said AI capital expenditures in 2026 will run between $115 billion and $135 billion, nearly twice last year's figure. That level of spending warrants a return the open-source playbook never produced.

Meta Introduces Muse Spark, Ushering in a New Generation of Intelligent AI Systems

Muse Spark is designed to go beyond conventional AI models by combining multimodal perception with advanced reasoning capabilities. In a decisive move to reshape the future of artificial intelligence, Meta has unveiled Muse Spark, a next-generation multimodal reasoning model developed by its Meta Superintelligence Labs. As the first model in the newly introduced Muse family, Muse Spark marks a foundational step in the company's long-term vision of building "personal superintelligence" -- AI systems capable of deeply contextual understanding, adaptive reasoning, and seamless integration into everyday digital life.

Meta Ditches Llama: Can Muse Spark Revamp AI Roadmap? - Meta Platforms (NASDAQ:META)

Meta Ditches Llama, Analyst Expects Muse Spark To Revamp AI Roadmap Meta Platforms Inc (NASDAQ:META) launched Muse Spark, a new AI model from its Superintelligence Labs, on Wednesday, which marks a shift away from the Llama series. The new LLM (large language model) capabilities have the potential to strengthen the company's product pipeline and ads performance, according to BofA Securities. The Analyst: Analyst Justin Post maintained a Buy rating and price target of $885. The Thesis: The company indicated that Muse Spark represents an initial step in its AI roadmap, "with multiple advanced models currently in development," Post said in the note. Check out other analyst stock ratings. Muse Spark is expected to replace Meta Platforms' existing Llama-based chatbots across WhatsApp, Instagram, Facebook and smart glasses in the coming weeks, he added. The model launch should enhance the company's "AI capabilities & positioning across both consumer and enterprise offerings," the analyst wrote. Integrating Muse Spark into Meta Platforms' core ads stack is likely to improve targeting and personalization, driving ROAS (returns on ad spend) gains for core ads business, he further stated. META Price Action: Shares of Meta Platforms had risen by 2.18% to $625.76 at the time of publication on Thursday. Image: Shutterstock Market News and Data brought to you by Benzinga APIs To add Benzinga News as your preferred source on Google, click here.

Meta introduces Muse Spark AI model with multimodal reasoning and multi-agent capabilities

Meta has introduced Muse Spark, a new AI model developed by Meta Superintelligence Labs. The model is part of a broader Muse series designed to advance multimodal understanding, reasoning, and agent-based workflows, as part of Meta's effort toward personal superintelligence. Muse Spark is positioned as a foundation model that supports text and image inputs, structured reasoning, tool integration, and multi-agent collaboration. It is integrated into Meta AI across the Meta AI app and meta.ai platform. Muse Spark is a multimodal AI model that processes both text and images within a unified system. It is designed to handle reasoning tasks while combining information from different input types. Key aspects include: The model is designed to be small and fast while handling complex tasks across domains such as science, mathematics, and health. It also represents a broader overhaul of Meta's AI stack, supported by infrastructure investments including the Hyperion data center. Muse Spark is developed using a multi-stage scaling approach: 1. Pretraining Pretraining establishes foundational capabilities in reasoning, multimodal understanding, and coding. Meta rebuilt its pretraining stack with improvements in architecture, optimization, and data curation. These changes allow the model to achieve comparable performance with over an order of magnitude less compute than Llama 4 Maverick, improving efficiency across training. 2. Reinforcement learning Reinforcement learning is applied post-training to refine outputs. Despite typical instability challenges, Muse Spark shows stable and predictable improvements, including: Test-time reasoning focuses on how the model processes queries before generating responses. Meta uses: In some evaluations, the model shows a transition from longer reasoning to compressed reasoning using fewer tokens, followed by improved performance through expanded solutions. Muse Spark powers updates across Meta AI, including new interaction modes and workflows. Meta AI can deploy multiple agents simultaneously. For example, in trip planning: These tasks are handled in parallel to improve response quality and speed. Users can interact with Meta AI using images. The system can: This capability is also expected to extend to AI glasses, enabling real-world perception. Muse Spark also supports visual coding, allowing generation of websites, dashboards, mini-games, and simulations from prompts. Muse Spark has undergone evaluation under Meta's Advanced AI Scaling Framework, covering threat models, behavioral alignment, and adversarial robustness. Key findings include: Third-party evaluation by Apollo Research found that the model demonstrated high awareness of evaluation scenarios, sometimes identifying "alignment traps." Meta noted that this does not impact deployment decisions but requires further study. Muse Spark is the first model in Meta's Muse series, with future development focused on scaling capabilities across models and platforms. Contemplating mode is rolling out gradually within meta.ai, with additional access expanding over time.

Meta Revives its AI Dream with Muse Spark AI Model

The reaction was instant as Meta stock jumped 6.5% though one cannot ascribe all of it to the launch of Muse's first member called Spark Sometimes arriving late on the scene of action delivers better outcomes as the last to arrive often catches the attention of everyone present. Ask celebrities who regularly make a late entry in order to get the arch lights on them. Looks like Meta Platforms has borrowed this technique to reveal their new AI model called Muse Spark last night. CEO Mark Zuckerberg took to Threads to reveal that the new model family called "Muse" that appeared to suggest Meta's continued presence in the crowded AI space made up of the likes of OpenAI, Anthropic, Gemini, Perplexity and now Microsoft and a few others. They're calling it the "first step" toward an "overhaul of (their) AI efforts." The reaction was instant as Meta stock jumped 6.5% though one cannot ascribe all of it to the launch of Muse's first member called Spark. And yet, the release has come at an opportune time. Barely a month ago, the New York Times claimed Meta had put off this launch as its new models were falling short of their rivals' performance. Muse Spark is now available on the web and the Meta AI app, is expected to improve over time. The company plans to roll out a "Contemplating" mode, which allows it to tackle more complex problems. Meta's model uses multiple AI agents at once to work on the same problem, which it says will generate faster results for its Contemplating mode. Our work so far has focused on establishing strong scientific and technical foundations to methodically scale larger and more capable models. Today we are sharing our first milestone: Muse, our new family of models, the post on X says. Spark, the first model in the Muse family, powers a new version of Meta AI that you can try today. It's a world-class assistant and particularly strong in areas related to personal superintelligence like visual understanding, health, social content, shopping, games, and more. You can try the new Meta AI in the app or at meta.ai, says Zuckerberg in the post. "Looking ahead, we plan to release increasingly advanced models that push the frontier of intelligence and capabilities, including new open source models. We are building products that don't just answer your questions but act as agents that do things for you," the post says. Given that Meta spent billions to acquire expertise in the form of a new AI team led by Alexandr Wang (formerly of Scale AI) for its Meta Superintelligence Labs, the arrival of Muse Spark should reignite the spark around Meta AI. The question though is where exactly does Meta want to put its AI expertise to work? Zuckerberg gives us a clue by noting that the new Meta AI would be "particularly strong" in areas such as "health, social content, shopping, games." So, enterprise AI is not on the cards for Meta Platforms. This is good news for Anthropic that seems to have monopolised the enterprise market with OpenAI being a distant second (though its CFO Sarah Friar thinks otherwise). If Meta can progress as claimed on the chatbot front and turbocharge its ad business, it may end up giving stiff competition to OpenAI's efforts to build an ad business around ChatGPT. However, there could be some privacy concerns around Muse Spark. Meta's recent push into health and the requirement of using an existing Meta account such as Facebook or Instagram to log into Muse Spark could raise a few eyebrows. Though the company does not explicitly say that a user's Facebook or Instagram account information would be put to use by AI, there is a remote likelihood of it happening when Meta trains its LLMs on public user data.

Meta Debuts New LLM Muse Spark | PYMNTS.com

By completing this form, you agree to receive marketing communications from PYMNTS and to the sharing of your information with our sponsor, if applicable, in accordance with our Privacy Policy and Terms and Conditions. Muse Spark, announced by the tech giant Wednesday (April 8), powers the Meta AI app and website, and will be rolling out to its social media platforms and artificial intelligence (AI) glasses in the weeks ahead. "We are on our way to personal superintelligence: an assistant that can help anyone, anywhere with the things that matter most to them," Meta said in its announcement Wednesday (April 8). With this launch, the company added, comes upgraded versions of the MetaAI app and meta.ai, with the AI now able to handle both quick answers to questions and provide help with complex problems that require strong reasoning. "You can switch between modes depending on the task, and Meta AI can launch multiple subagents in parallel to tackle your question," the company said. "Like planning a family trip to Florida where one agent drafts the itinerary, another compares Orlando vs. the Keys, and a third finds kid-friendly activities -- all at the same time, giving you a better answer, faster." Muse Spark also now lets Meta AI users get more detailed responses to health questions, including those that involve images and charts, the announcement added. "Health is one of the top reasons people turn to AI, so we worked with a team of physicians to develop the model's ability to provide helpful information on common health questions and concerns," the company said. In addition, the model "excels at visual coding," allowing users to create custom websites and mini-games straight from a prompt. Unlike past Meta AI models, this one is close-sourced, though the company says it hopes to open-source future versions. The launch follows a year in which Meta spent billions to court AI talent as it competes with the likes of OpenAI and Google in the artificial intelligence space. Writing about Meta's AI projects last year, PYMNTS contrasted the company's efforts with those of competitors Amazon, Google and Microsoft. While those companies pair their AI investments with clear paths to revenue, Meta had focused on building for internal use. The report pointed out the example of Microsoft monetizing OpenAI's models through Copilot/Azure subscriptions, Google offering Gemini and Vertex AI access through its cloud unit, and Amazon deriving enterprise revenue from the Bedrock and SageMaker platforms. "Meta, by contrast, uses its AI for engagement, recommendation engines, ad ranking, and tools like Meta AI and Reels," PYMNTS wrote. "Those may improve user metrics, but it's not clear how they will contribute to the bottom line."

Meta unveils first AI model from superintelligence team

Meta Platforms on Wednesday unveiled Muse Spark, the first artificial intelligence model from a team it assembled last year through a costly talent war and sweeping internal restructuring to catch up with rivals in the AI ⁠race. Meta Platforms on Wednesday unveiled Muse Spark, the first artificial intelligence model from a team it assembled last year through a costly talent war and sweeping internal restructuring to catch up with rivals in the AI ⁠race. US ⁠tech giants are under pressure to prove their massive AI outlays will pay off. The stakes are especially high for Meta after it hired Scale AI CEO Alex Wang last year under a $14.3 billion deal and offered some engineers ⁠pay packages of hundreds of millions of dollars to staff a new superintelligence team. Muse ⁠Spark is the first in a new series of models from that team, which is pursuing machines that can outthink humans. It will initially be available only on the lightly-used Meta AI app and website, and in the coming weeks, replace the existing Llama models powering chatbots on WhatsApp, Instagram, Facebook and Meta's collection of smart glasses. "This initial model is small ⁠and fast by design, yet capable enough to reason through complex questions in science, math, and health. It is a powerful foundation, and the next generation is already in development," the company said in a blog post.

Meta Bets Big On Superintelligence With Muse Spark Launch -- Analyst Sees 'Significant' Monetization Potent

The AI model, called Muse Spark, initially code-named Avocado, is the first in the new Muse series. Meta Superintelligence Labs has rebuilt the AI stack from scratch in the past nine months, marking a faster development cycle than any previous ones. Muse Spark Powers AI, Shopping, And API Meta is positioning Muse Spark not as a top-tier model but is highlighting its efficiency and competitive performance across various tasks. The company plans to open-source future versions of the model, a shift from its current proprietary stance. "Muse Spark offers competitive performance in multimodal perception, reasoning, health, and agentic tasks," Meta said. The Mark Zuckerberg-led company is aiming to create a new revenue stream by providing third-party developers with access to Muse Spark's core technology through an API. This model currently powers the company's digital assistant on both the standalone Meta AI app and its desktop website. Muse Spark will include a Shopping mode that helps users buy clothes or decorate rooms by leveraging inspiration and ideas from creators and communities on its apps. The new model will soon be integrated into Facebook, Instagram, WhatsApp, Messenger, and the company's Ray-Ban Meta AI glasses. It is also expected to power the company's Vibes AI video feature in the Meta AI app. Meta Doubles Down On AI Despite Delays Benzinga's Edge Rankings place Meta in the 89th percentile for quality and the 84th percentile for growth, reflecting its strong performance in both areas. Benzinga's screener allows you to compare Meta's performance with its peers. META Price Action: On a year-to-date basis, Meta declined 5.84%, as per Benzinga Pro. On Wednesday, the stock climbed 6.50% to close at $612.42. Disclaimer: This content was partially produced with the help of AI tools and was reviewed and published by Benzinga editors. Image via Shutterstock Market News and Data brought to you by Benzinga APIs To add Benzinga News as your preferred source on Google, click here.

Meta rolls out new AI model in 'fundamental shift' in latest effort to catch up with rivals

Mark Zuckerberg-run Meta has rolled out a new artificial intelligence model designed to power everything from shopping suggestions to chat -- the latest effort in the tech giant's costly push to catch up in the AI race. The company earlier this week unveiled Muse Spark, its first big artificial intelligence model since Meta overhauled its internal AI division in a bid to close the gap with rivals like OpenAI and Google. The latest push comes after Meta poured billions into the Scale AI startup -- whose founder Alexandr Wang reportedly went on to describe Zuckerberg's micromanagement as "suffocating." Muse Spark is designed to handle text, images and more complex reasoning tasks, allowing users to ask questions, analyze photos, generate content and even get help with shopping decisions. Ravi Sawhney, founder of RKS Design, said Meta's push into AI shopping is less about technology, and more about influencing behavior. "Meta is trying to move shopping from intent to influence. Instead of people searching for what they want, the platform is shaping what they believe they want in real time," he told The Post. "That is a fundamental shift." Zuck is increasingly tying AI directly to consumer products rather than focusing solely on developer tools or open-source releases. Meta is baking the technology into new "shopping mode" features that suggest products, compare items and surface recommendations based on what users are already browsing across its apps. The company has pitched the assistant as more like a personal aide than a chatbot -- capable of handling everyday decisions like what to wear, how to decorate a room or which products to buy. "The opportunity is not better recommendations. It is creating a sense of confidence and self alignment in the decision," Sawhney told The Post. "Most AI shopping tools will fail here. They will surface more options, more noise and more second guessing." He framed the strategy as a direct challenge to existing tech giants. "Amazon wins on intent. Google wins on information. Meta is betting it can win on identity and discovery," he told The Post. "That only works if users trust what they are being shown and feel understood, not managed." The flashy new features come as Meta scrambles to regain ground in a fast-moving AI arms race that it once helped shape. The company said the model is already live in its Meta AI app and website, with plans to expand it across Facebook, Instagram, WhatsApp, Messenger and its Ray-Ban smart glasses in the coming weeks. The Facebook parent had been viewed as a leader in open-source AI with its Llama models, but its most recent releases failed to impress compared with offerings from competitors like OpenAI's ChatGPT and Anthropic's Claude. That prompted a sweeping reset effort inside the company. Last year, Meta reorganized its AI efforts under a new division called Meta Superintelligence Labs and embarked on an aggressive hiring spree, recruiting researchers and executives from top rivals. The centerpiece of that effort was a $14.3 billion investment in Scale AI -- a deal widely seen as a way to bring Wang, into a leadership role overseeing Meta's AI strategy. In December, a Financial Times report cited sources in the know as saying that Wang was chaffing under Zuckerberg. Meta has also snapped up teams from startups and rival labs while reshuffling its internal structure multiple times in recent months in an attempt to speed up development. At the same time, Meta has been ramping up spending on the infrastructure needed to build and run advanced AI systems -- committing tens of billions of dollars to data centers, chips and cloud capacity. The company told investors it expects to spend as much as $135 billion on AI-related capital expenditures this year alone, underscoring the scale of its bet. Shares of Meta rose more than 3% in the afternoon trading session on Wall Street on Thursday. As of around 1:30 p.m. ET, the stock was trading at around $633 per share.

Meta Platforms surges 7% amid AI model debut By Investing.com

Investing.com -- Meta Platforms (NASDAQ:META) shares rose 7% Wednesday following the company's introduction of Muse Spark, its first artificial intelligence model release since Chief Executive Officer Mark Zuckerberg initiated a multibillion-dollar reorganization of its AI operations. The stock gained amid broad strength in technology shares and the debut of the new AI model, which represents the initial offering in Meta's Muse family developed by Meta Superintelligence Labs. Muse Spark features native multimodal reasoning capabilities, including tool-use support, visual chain of thought, and multi-agent orchestration. The model is available at meta.ai and through the Meta AI app, with a private API preview opening to select users. Meta also introduced Contemplating mode, which operates multiple agents that reason simultaneously, achieving 58% performance in Humanity's Last Exam and 38% in FrontierScience Research. The company rebuilt its pretraining stack over nine months with improvements to model architecture, optimization, and data curation. Meta reports reaching the same capabilities with over an order of magnitude less compute compared to its previous model, Llama 4 Maverick. The model's applications include multimodal visual integration across domains and health reasoning capabilities. Meta collaborated with over 1,000 physicians to develop training data for health-related responses. Muse Spark can generate interactive displays explaining nutritional content and muscles activated during exercise. In safety evaluations conducted under Meta's Advanced AI Scaling Framework, Muse Spark demonstrated strong refusal behavior across high-risk domains including biological and chemical weapons. Third-party evaluations by Apollo Research found the model showed the highest rate of evaluation awareness among models they have observed.

Meta reignites AI ambitions with Muse Spark

Until now, Meta had primarily relied on its open-source Llama model to establish itself in the AI space. However, faced with performance levels deemed insufficient, the group is shifting gears and introducing Muse Spark, the first model from its new Meta Superintelligence Labs entity, led by Alexandr Wang, founder of Scale AI. A model designed for reasoning and practical application Muse Spark distinguishes itself as a native multimodal reasoning model. It supports tool usage, visual reasoning, and multi-agent orchestration. Already integrated into the Meta AI application and the recently updated meta.ai website, it will be progressively deployed across WhatsApp, Instagram, Facebook, Messenger, and the group's smart glasses. Meta presents this model as both compact and fast, while remaining capable of processing complex problems in fields such as science, mathematics, and healthcare. The group specifies that this is a first step, with several more advanced models currently under development. Improvements in architecture, optimization, and data selection are expected to allow Muse Spark to outperform existing models in terms of efficiency. Renewed credibility among analysts The reaction from analysts has been broadly positive. Bank of America believes these new capabilities could strengthen Meta's entire ecosystem, particularly on the advertising front. "The integration of Muse Spark into Meta's core advertising platform should improve targeting and personalization, thereby optimizing return on investment," the group noted. Some observers had previously pointed to a delay in the model's development, with a launch initially expected in May. For Bank of America, this early release helps dispel uncertainties weighing on the stock, while highlighting an ambitious roadmap to enhance model performance in the coming months. Muse Spark could thus help reposition Meta in the competitive landscape, much like what Gemini 3 represented for Alphabet. JP Morgan shared a similar sentiment. Doug Anmuth noted that investor expectations had been revised downward in recent months, particularly following disappointing results and doubts regarding the group's AI progress. In this context, the launch of Muse Spark appears to be a factor capable of restoring confidence in Meta's growth strategy and improving market sentiment.

Meta launches Muse Spark to reboot its artificial intelligence strategy

Meta has introduced Muse Spark, a new artificial intelligence model developed by Meta Superintelligence Labs, led by Alexandr Wang. This launch follows the mixed reception of Llama 4 and reflects a drive to regain a leading position against rivals such as OpenAI, Google, and Anthropic. The group is also pivoting its strategy toward proprietary models, while keeping the possibility of releasing certain versions open. According to the company, Muse Spark is built on a completely redesigned infrastructure and optimized training methods. The model distinguishes itself by delivering performance comparable to much larger systems while significantly reducing computational requirements. It is designed to handle complex tasks, particularly in scientific, mathematical, and medical fields, featuring advanced reasoning and multimodal understanding capabilities. Meta acknowledges, however, that there is still room for improvement, particularly regarding long-duration autonomous systems and software development applications. This launch comes amid massive investment in AI infrastructure, where execution speed and technological efficiency have become central challenges for industry players.

Meta Platforms Rises on Release of New AI Model

Shares of Meta Platforms gained after the company announced a new large language artificial-intelligence model. The stock was up 6.6%, at $613.05, midday Wednesday, but has fallen about 7.1% year-to-date. The model, Muse Spark, is the first of an upcoming family of models developed by Meta Superintelligence Labs, the company said in a Wednesday blog post. Meta has poured billions of dollars to hire AI talent in a push to catch up to frontier AI labs such as OpenAI, Anthropic, and Google DeepMind, according to The Wall Street Journal. "Muse Spark is the first step on our scaling ladder and the first product of a ground-up overhaul of our AI efforts," Meta said. The company said it is opening a private application programming interface preview to certain users. Meta may open-source some versions of the model, but doesn't have plans to widely release its API, the Journal reported. The model outscored Google Gemini on some internal tests and was competitive with OpenAI and Anthropic models on others, while significantly beating xAI's Grok on many benchmarks, Meta said.

Meta unveils first AI model from costly superintelligence team

April 8 (Reuters) - Meta Platforms on Wednesday unveiled Muse Spark, the first artificial intelligence model from a costly team it assembled last year to catch up with rivals in the AI race. Shares of the company extended gains to trade up nearly 7%. U.S. tech giants are under pressure to prove their massive AI outlays will pay off. The stakes are especially high for Meta after it hired Scale AI CEO Alex Wang last year under a $14.3 billion deal and offered some engineers pay packages of hundreds of millions of dollars to staff a new superintelligence team. Superintelligence refers to AI machines that could outthink humans. Muse Spark is the first in a new series of models from that team, and is part of a family of models known internally as Avocado. The model will initially be available only on the lightly-used Meta AI app and website, and in the coming weeks, replace the existing Llama models powering chatbots on WhatsApp, Instagram, Facebook and Meta's collection of smart glasses. "This initial model is small and fast by design, yet capable enough to reason through complex questions in science, math, and health. It is a powerful foundation, and the next generation is already in development," the company said in a blog post. It did not disclose the model's size, a key measure typically used to compare an AI system's computing power with rivals. Meta CEO Mark Zuckerberg had tempered expectations for performance, telling investors in January that he thought the team's first models "will be good but, more importantly, will show the rapid trajectory that we're on." "I expect us to steadily push the frontier over the course of the year as we continue to release new models," he had said. FOCUS ON EVERYDAY TASKS Muse Spark can help users with tasks such as estimating the calories in a meal from a photo or superimposing an image of a mug on a shelf to see how it looks. Meta also released Contemplating mode, which runs multiple AI agents in parallel to boost reasoning power, allowing Muse Spark to take on the extended thinking modes of Google's Gemini Deep Think and OpenAI's GPT Pro. The company is betting that applying superintelligence to everyday personal tasks will help it tap its more than 3.5 billion users across its social media platforms, potentially giving it an edge over rivals with a smaller reach. (Reporting by Aditya Soni in Bengaluru and Katie Paul in New York; Editing by Leroy Leo)

Meta unveils first AI model from superintelligence team

April 8 (Reuters) - Meta Platforms on Wednesday unveiled Muse Spark, the first artificial intelligence model from a team it assembled last year through a costly talent war and sweeping internal restructuring to catch up with rivals in the AI race. U.S. tech giants are under pressure to prove their massive AI outlays will pay off. The stakes are especially high for Meta after it hired Scale AI CEO Alex Wang last year under a $14.3 billion deal and offered some engineers pay packages of hundreds of millions of dollars to staff a new superintelligence team. Muse Spark is the first in a new series of models from that team, which is pursuing machines that can outthink humans. It will initially be available only on the lightly-used Meta AI app and website, and in the coming weeks, replace the existing Llama models powering chatbots on WhatsApp, Instagram, Facebook and Meta's collection of smart glasses. "This initial model is small and fast by design, yet capable enough to reason through complex questions in science, math, and health. It is a powerful foundation, and the next generation is already in development," the company said in a blog post. (Reporting by Aditya Soni in Bengaluru and Katie Paul in New York; Editing by Leroy Leo)

Meta launches Muse Spark AI: What it is and what it can do

According to Meta, Muse Spark is 'the first step toward a personal superintelligence that understands your world.' Meta has introduced a new AI model dubbed Muse Spark. It is the first model in its Muse family and has been developed by Meta Superintelligence Labs. Muse Spark is a natively multimodal reasoning model which supports tool-use, visual chain of thought and multi-agent orchestration. According to Meta, the model is part of a larger effort to scale its AI capabilities. 'Muse Spark is the first step on our scaling ladder and the first product of a ground-up overhaul of our AI efforts,' the company said. Meta also noted that it is investing across its entire AI stack, including research, training systems and infrastructure. Also read: Anthropic launches Project Glasswing to fight AI-driven cyberattacks, know how Muse Spark is a multimodal reasoning model. According to Meta, it is 'the first step toward a personal superintelligence that understands your world.' One of its key features is something Meta calls Contemplating mode. This mode allows multiple AI agents to reason about a problem in parallel before delivering an answer. The company claims this feature allows the model to compete with the extreme reasoning modes of frontier models such as Gemini Deep Think and GPT Pro. Muse Spark can analyse visual information, recognise objects and help answer complex STEM questions. Health is another major focus. Meta says it collaborated with more than 1,000 physicians to create training data that improves the model's ability to answer health-related questions. Also read: Sam Altman misled board on GPT-4 safety approvals before getting fired, claims report Meta's Muse Spark AI model is currently available through Meta AI on meta.ai and the Meta AI app. The company has also opened a private API preview for select users. Meta says the Contemplating mode feature will roll out gradually.