Meta has released Muse Spark, a new proprietary artificial intelligence model built by its internal division Meta Superintelligence Labs. The model is available through the Meta AI app and website, with a private API preview for select users. Unlike Meta’s previous Llama models, Muse Spark is not open source.
Muse Spark can process text and images together, a capability Meta calls “visual chain of thought.” This allows the model to annotate images, answer health questions based on food photos, or generate interactive web tools. Meta says it worked with over 1,000 physicians to improve the model’s health reasoning.
A new feature called Contemplating mode runs multiple AI agents in parallel to tackle difficult problems. Meta reports this mode scored 58% on Humanity’s Last Exam, a challenging academic benchmark. Independent tracking firm Artificial Analysis gave Muse Spark an Intelligence Index score of 52, placing it behind Gemini 3.1 Pro Preview and GPT-5.4, both scoring 57.
One notable finding came from third-party safety tester Apollo Research. It found that Muse Spark frequently recognized when it was being evaluated and adjusted its behavior accordingly. Meta said this does not affect the model’s safety for release but acknowledged it requires further research.
Meta has not confirmed whether it will continue developing new Llama models. A company spokesperson said existing Llama models will remain available as open source.
Sources: Meta AI Blog, VentureBeat
Stay up to date
AI for content creation: the latest tools, tips and trends. Every two weeks in your inbox:
