Editorial illustration for Meta Announces First Model From Its New AI Lab, Muse Spark
Meta Debuts Muse Wonder From New AI Lab
Meta released two new AI models on Friday, Muse Wonder and Muse Wonder Full, the first products out of Meta Superintelligence Labs since Mark Zuckerberg formed the unit last year. The company says Muse Wonder Full has "frontier capabilities," its term for ranking among the world's most capable models. Both will ship through an API and through Muse Code, a coding tool, with the top-tier version also set to power Meta's Muse personal agent and other features across the company's apps.
The release caps a turbulent stretch for Meta's AI ambitions. Zuckerberg poured billions into Meta Superintelligence Labs and brought in Alexandr Wang, formerly CEO of Scale AI, to run it, a bet that drew plenty of skepticism given staff turnover inside the lab. Meta built its reputation in AI on the Llama series, once the first frontier open-weight models from a major lab and a favorite among researchers and startups for being downloadable and runnable on personal hardware.
That edge eroded as later Llama releases underwhelmed while Anthropic, OpenAI, and Google kept pushing out stronger models. Muse Wonder is Meta's attempt to close that gap.
Muse Wonder was built to compete with the very best, however. Meta says the largest version of the new model outperforms Anthropic’s second-best model, Claude Fable 5.1, and ranks just below its best, Claude Opus 5.5, on the Artificial Analysis Intelligence Index, a widely used aggregated score of model capabilities.
Why this matters Meta's framing here is confident but light on substance. Since April, when Muse Spark debuted, the company has rolled out a small open-weight model, Muse Glimmer, and now two more entries, Muse Wonder and Muse Wonder Full, without saying how either was built. That gap matters for anyone trying to evaluate the "frontier" claim rather than just accept it.
Developers and researchers can't replicate or stress-test what they can't see, and Meta's promise of "larger open-weight models in the future" is still just a promise, not a release date. For founders weighing which lab to build on, the pattern is worth tracking: Meta talks in terms of a "scaling ladder" and leaps forward, but the actual weights, training data, and benchmarks that would let outsiders verify those leaps stay closed. We'd treat Zuckerberg's pivot as a real bet that's generating real models, but the lack of technical disclosure means the frontier claim rests on Meta's word alone for now.
Watch whether Glimmer's open-weight successor actually ships, and on what terms.
Related Reading
Browse all Meta news →
Business & Startups Nordic pilot adds Gemini for Education, NotebookLM to boost AI literacy
Business & Startups Kling launches Video O1, all-in-one model with MVL bridge using transformer
Business & Startups DeepSeek Seeks More Capital Weeks After USD 7B Funding Round
Related Topic Meta’s Muse Tracks Your Email and Bank Data While You Work
AI Tools & Apps
Related Topic Meta Faces Lawsuit Over AI Training Data, Facial Recognition
Policy & Regulation
From Archives Microsoft's Andreou: OpenAI Models Excel in Specific Tasks
From Archives Odyssey-3 Demo Lets You Generate Interactive Worlds From Text
From Archives Meta's AI Logo Designer Faces Backlash After Revealing Workshop Fee
AI Tools & Apps
From Archives Meta Rolls Out AI Tools to Detect Ads Leading to Child Abuse Material
AI Tools & Apps
Common Questions Answered
What are the two new AI models that Meta released from its Superintelligence Labs?
Meta released Muse Wonder and Muse Wonder Full, marking the first products from Meta Superintelligence Labs since Mark Zuckerberg formed the unit last year. Muse Wonder Full is described by Meta as having "frontier capabilities," positioning it among the world's most capable AI models.
How does Muse Wonder's performance compare to Anthropic's Claude models?
According to Meta, the largest version of Muse Wonder outperforms Anthropic's Claude Fable 5.1 and ranks just below Claude Opus 5.5 on the Artificial Analysis Intelligence Index, a widely used aggregated score of model capabilities. This positioning places Muse Wonder among the top-tier AI models currently available.
Through which platforms will Meta's new Muse models be available?
Both Muse Wonder and Muse Wonder Full will be shipped through an API and through Muse Code, a coding tool developed by Meta. Additionally, the top-tier version will power Meta's Muse personal agent and other features across the company's apps.
What concern does the article raise about Meta's transparency regarding the Muse models?
The article notes that Meta has not disclosed how either Muse Wonder or Muse Wonder Full were built, creating a gap that makes it difficult for developers and researchers to evaluate Meta's "frontier" claims rather than simply accepting them. Without visibility into the models' construction, developers cannot replicate or stress-test these systems to verify their capabilities.
What other Muse models has Meta released since April?
Since the debut of Muse Spark in April, Meta has rolled out Muse Glimmer, a small open-weight model, in addition to the newly announced Muse Wonder and Muse Wonder Full. This represents Meta's expanding portfolio of AI models from its Superintelligence Labs.
Further Reading
- Introducing Muse Spark: Meta's Most Powerful Model Yet - Meta
- Introducing Muse Spark: Scaling Towards Personal Superintelligence - AI at Meta
- Introducing Muse Image and Muse Video - AI at Meta
- Introducing Muse Spark 1.1 - AI at Meta
- Meta AI Research: Muse Spark, Muse Glimmer and Muse Image - AI at Meta
