Meta debuts Muse Spark
Meta released Muse Spark, its first model from the superintelligence team, and has already started routing queries through the Meta AI app and Meta.ai site. (reuters.com) Muse Spark accepts voice, text and image inputs but currently returns text-only output, and Meta plans to expand the model across Facebook, Instagram and WhatsApp. (axios.com)
Meta just swapped the engine inside its chatbot before most people noticed the hood was open. On April 8, Meta said its new Muse Spark model is already powering the Meta AI app and Meta.ai, making this a live product change, not a lab demo. (about.fb.com) This is the first model from Meta Superintelligence Labs, the team Meta built after a costly hiring spree and internal reshuffle aimed at catching OpenAI, Google, and Anthropic. Reuters said the push included Meta’s $14.3 billion Scale AI deal and compensation packages worth hundreds of millions of dollars for some engineers. (money.usnews.com) Meta says Muse Spark was built over the past nine months and was internally code-named Avocado. Axios reported that the model is the first big output from the group led by Alexandr Wang, the former Scale AI chief who joined Meta last year. (axios.com) The immediate change is not that Meta built a chatbot from scratch. The immediate change is that Meta is replacing the old Llama family inside its assistant with a new in-house line called Muse, starting with Spark. (money.usnews.com) (about.fb.com) Meta is pitching Spark as small and fast rather than as the biggest model on the market. In its launch post, the company said the model was designed to handle harder reasoning questions in science, math, and health while still responding quickly enough for consumer apps. (about.fb.com) Meta also says the assistant can now split one request into several parallel tasks, like having multiple researchers work the same assignment at once. In Meta’s example, one agent drafts a Florida trip plan while others compare Orlando with the Florida Keys and look for child-friendly activities. (about.fb.com) Spark can take in text, voice, and images, which means you can talk to it, type to it, or point a camera at something and ask a question about what it sees. Meta’s own example is photographing an airport snack shelf so the assistant can identify which items have the most protein. (about.fb.com) (axios.com) The bigger bet is distribution. Meta said Spark will roll out in the coming weeks to WhatsApp, Instagram, Facebook, Messenger, and its artificial intelligence glasses, which gives one model a path into apps used by billions of people. (about.fb.com) (money.usnews.com) There is also a strategy shift hiding inside the launch. Bloomberg reported that Muse Spark is a closed model, which breaks from Meta’s earlier habit of making its flagship Llama systems openly available, and Meta said it will offer Spark in private preview through an application programming interface to selected partners instead of releasing it broadly. (bloomberg.com) (about.fb.com) That helps explain why this launch feels more defensive than triumphant. CNBC reported that Meta is trying to regain momentum after the weak reception to Llama 4, and Muse Spark is the company’s first public proof that the expensive Wang-led rebuild produced something Meta is willing to ship across its consumer products. (cnbc.com)