Meta launches Muse Spark
Meta unveiled Muse Spark, its first major large language model from the newly formed Superintelligence Labs, signaling a shift from the company's earlier open-release approach. Reports frame this as a move toward more proprietary, product-focused models and the market reacted positively with a noticeable share-price bump after the announcement. (cnbc.com) (invezz.com)
Meta spent years telling the world that its big artificial intelligence bet was Llama, a family of models it released openly. On April 8, it changed course with Muse Spark, a model Meta says will stay mostly inside its own products instead of being broadly released. (cnbc.com) (about.fb.com) Muse Spark is the first major model from Meta Superintelligence Labs, the new group led by chief artificial intelligence officer Alexandr Wang. Reuters reported that Meta built that team last year to catch up with OpenAI and Google after frustration with the pace of its earlier Llama work. (reuters.com) (techcrunch.com) A large language model is the text engine behind a chatbot. It is the part that predicts the next word, writes answers, and follows instructions, like a very expensive autocomplete trained on huge piles of text and images. (about.fb.com) (cnbc.com) Meta says Muse Spark is a multimodal reasoning model. That means it can work across text, images, and other inputs while handling longer chains of questions, the kind of tasks that make assistants feel less like search boxes and more like software that can plan. (about.fb.com) (siliconangle.com) The company says it rebuilt its artificial intelligence stack over the last nine months and used Muse Spark as the first proof that the new setup works. Meta also says the model reaches its reasoning performance with more than an order of magnitude less compute than Llama 4 Maverick, its previous mid-size flagship. (about.fb.com) (venturebeat.com) The distribution plan is the real break from the old Meta playbook. Meta says Muse Spark already powers the Meta AI app and website, and will roll out to WhatsApp, Instagram, Facebook, Messenger, and its artificial intelligence glasses in the coming weeks. (about.fb.com) (tbreak.com) That is different from how Meta handled Llama. In 2024 and 2025, Meta pushed Llama as an open model family for developers, while Muse Spark is being kept largely inside Meta’s own ecosystem, with reports describing only a limited private application programming interface preview for selected partners. (about.fb.com) (aol.com) (tbreak.com) Investors liked the switch. CNBC reported that Meta shares rose after the announcement, and other market coverage put the jump at roughly 7 percent to 9 percent on April 8 as traders read Muse Spark as evidence that Meta may finally have a cleaner product story in artificial intelligence. (cnbc.com) (invezz.com) (blockonomi.com) The backdrop is a spending race that has gotten very expensive very fast. Meta has been signing infrastructure deals with Advanced Micro Devices, Nvidia, Arm, and Corning while pitching 2026 as a year when artificial intelligence would drive performance across the company. (about.fb.com 1) (about.fb.com 2) So Muse Spark is not just one more chatbot model. It is Meta saying that the next phase of its artificial intelligence business will be less about giving away the engine and more about using that engine inside Facebook, Instagram, WhatsApp, Messenger, and glasses that it controls end to end. (about.fb.com) (reuters.com)
