Meta rolls out Muse Spark

Meta has moved from open-weight models to a proprietary model called Muse Spark, which the company says will power Meta AI across Facebook, Instagram, WhatsApp and Messenger. Reports say Muse Spark was built in nine months and includes a new shopping mode aimed at tailoring product suggestions for creators and brands. (buildfastwithai.com, sangritoday.com)

Meta has started replacing the Llama models behind its assistant with a new in-house system called Muse Spark, a proprietary model Meta says will power Meta AI across its apps. (about.fb.com) Meta announced Muse Spark on April 8, 2026, and said it already powers the Meta AI app and website. The company said it will roll out to WhatsApp, Instagram, Facebook, Messenger and its artificial intelligence glasses in the coming weeks. (about.fb.com) Reuters reported Muse Spark is the first model from Meta Superintelligence Labs, the team Meta assembled last year after a broad internal shake-up and an expensive hiring push. Reuters also reported the model will replace the existing Llama systems that currently run chatbots across Meta’s consumer products. (reuters.com) Muse Spark is a closed model, which means Meta is no longer leading this product push with open-weight releases like Llama that outsiders could download and modify. Meta said Muse Spark was built for its own products and will unlock features that cite recommendations and content people share across Instagram, Facebook and Threads. (about.fb.com) That marks a change from the company’s recent artificial intelligence strategy. Llama helped Meta argue that open models could spread its technology widely, but Muse Spark is being positioned first as the engine for Meta’s own assistant, where speed, personalization and product integration matter more than outside access. (techcrunch.com) Meta said the model is “small and fast by design” and described it as capable of handling questions in science, math and health. TechCrunch reported Meta is also preparing a “Contemplating” mode that uses multiple agents working on the same problem to spend more time reasoning without slowing responses as much. (about.fb.com, techcrunch.com) Axios reported the project was code-named Avocado and built over nine months by a team led by Alexandr Wang. TechCrunch reported Wang, the former Scale AI chief executive, now leads Meta Superintelligence Labs after Meta invested $14.3 billion in Scale AI for a 49 percent stake. (axios.com, techcrunch.com) Meta’s own announcement emphasized personalization and product tie-ins, not just benchmark scores. The company said Muse Spark will eventually support features that surface recommendations and shared content from its social apps, a sign that Meta is tying its assistant more tightly to its advertising, creator and commerce ecosystem. (about.fb.com) Outside coverage has described new commerce features around the launch, including shopping tools aimed at matching products to users, creators and brands, but Meta’s April 8 announcement did not spell out a separate “shopping mode” by that name. What Meta did confirm is that Muse Spark is being built directly into consumer products rather than released mainly as a model for developers. (about.fb.com, buildfastwithai.com) The immediate test is whether people notice better answers inside apps they already use. After years of promoting Llama as Meta’s artificial intelligence banner, the company is now asking billions of users to meet Muse Spark first as a feature, not as a download. (about.fb.com, reuters.com)

Get your own daily briefing

Scout delivers personalized news, insights, and conversations tailored to your role and industry.

Download on the App Store

Shared from Scout - Be the smartest in the room.