Late last year, news emerged that Mark Zuckerberg\u2019s Meta would be shedding its open source roots to instead work on a closed model like the vast majority of its competitors.<\/p>\n
Now we\u2019ve finally gotten a first glimpse of the fruit of its labor: Muse Spark, codenamed Avocado and developed by the company\u2019s unbelievably expensive Superintelligence Labs<\/a>.<\/p>\n But there\u2019s a big problem that could undermine its flashy new announcement. Despite investors buying into the enthusiasm, sending Meta\u2019s shares soaring six percent following the announcement, the company admitted it likely won\u2019t be able to keep up with competing models.<\/p>\n An executive told Bloomberg<\/a> that the new model won\u2019t be able to keep up with OpenAI\u2019s ChatGPT, Anthropic\u2019s Claude or Google\u2019s Gemini. In a blog post announcing the new model<\/a>, the company admitted Muse Spark \u201cis an early data point on our trajectory, and we have larger models in development.<\/p>\n As such, the announcement is a bit of an enigma: if it can\u2019t keep up with the competition, why release it at all?<\/p>\n There\u2019s a good change Meta is just trying to get its foot in the door \u2014 or a \u201cseat at the big kid\u2019s table,\u201d as Wired put it<\/a>. The company has struggled to stay relevant in a rapidly changing landscape, making headlines for being found liable<\/a> in court for getting underage users dangerously addicted to social media last month instead of its AI efforts.<\/p>\n The company\u2019s 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 \u201cdistillation<\/a>,\u201d or training a \u201cstudent\u201d model on a more capable \u201cparent\u201d one, has proven controversial in the past<\/a>.<\/p>\n Meta\u2019s 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.<\/p>\n The results of the model, which flopped after being released almost exactly a year ago, \u201cwere fudged a little bit,\u201d as former Meta AI head Yann LeCun, who left the company amid the drama, told the Financial Times<\/a> in January.<\/p>\n \u201cMark was really upset and basically lost confidence in everyone who was involved in this,\u201d he added at the time. \u201cAnd so basically sidelined the entire GenAI organization. A lot of people have left, a lot of people who haven\u2019t yet left will leave.\u201d<\/p>\n Since its disastrous AI model launch, the company went on a hiring spree<\/a>, spending untold hundreds of millions<\/a> of dollars on top AI talent in an effort to scrounge together a Superintelligence Labs team capable of putting Meta back in the game.<\/p>\n But plenty of questions remain whether Muse Spark will be able to jumpstart Zuckerberg\u2019s 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.<\/p>\n Other early benchmarks paint a more flattering picture of Muse Sparks, with AI benchmarking company Artificial Analysis finding its score places it \u201cwithin the top 5 models we have benchmarked<\/a>.\u201d<\/p>\n Yet that doesn\u2019t hide the reality that Meta still has plenty of catching up to do. Besides, an AI model purportedly capable of telling how many calories<\/a> there are in a cup of white rice or \u201cplanning a family trip to Florida\u201d doesn\u2019t exactly feel like the cutting edge in 2026.<\/p>\n For now, Meta\u2019s Muse Spark will be free for all users. However, the company executive told Bloomberg that it\u2019s considering paywalling it behind a subscription in the future.<\/p>\n