April 8 - Meta Platforms on Wednesday introduced Muse Spark, the first AI model produced by a high-cost superintelligence team assembled last year to narrow the gap with competing technology firms. The company said the model is designed to be small and fast while still able to reason through complex questions in areas such as science, math and health.
Muse Spark will initially be available only on the lightly used Meta AI app and the company’s website. Meta said that in the coming weeks the model will replace the Llama models currently powering chatbots across WhatsApp, Instagram, Facebook and its line of smart glasses.
In a company blog post, Meta described Muse Spark as a foundational model and said a next-generation version is already in development. The post did not disclose the model’s size - a metric often used to compare the computing scale of AI systems.
Muse Spark belongs to a broader family of models that Meta refers to internally as Avocado. The company highlighted example user-facing capabilities such as estimating the calories in a meal from a photograph and superimposing an image of a mug on a shelf to preview how it would look. Meta acknowledged that some rivals already provide similar features.
Alongside Muse Spark, Meta released what it calls Contemplating mode - a function that runs multiple AI agents in parallel to increase reasoning power. The company framed Contemplating mode as a tool that enables Muse Spark to address extended thinking tasks that companies have been pursuing in other advanced models.
Meta has been under pressure to demonstrate that its large investments in artificial intelligence will generate returns. The stakes for the initiative have been heightened by significant personnel and financial commitments made last year, including the hiring of Scale AI CEO Alex Wang under a reported $14.3 billion deal and pay packages that reportedly reached into the hundreds of millions of dollars for some engineers to staff the new superintelligence team.
Meta said it believes applying superintelligence to everyday personal tasks could leverage its reach across more than 3.5 billion users on its social media platforms, a potential advantage compared with competitors that have smaller user footprints.
The company did not provide additional technical metrics for Muse Spark in its announcement, and it framed the release as the opening entry in a sequence of models to come from the team focused on more advanced machine reasoning.