Meta has launched its first frontier-class artificial intelligence models. The company released Muse Wonder and Muse Wonder Full on 12 October 2026. Frontier-class means the models aim to match the best AI systems in the world.
The name Muse Wonder comes from Meta’s own AI assistant, Muse. The company has spent heavily on AI talent and data centres over the past two years. Investors have been waiting to see real products come out of that spending.
Meta says Muse Wonder closes the gap with OpenAI and Anthropic on coding, finance and legal tasks. The models are available through the Meta Model API and inside Muse Code, Meta’s coding tool.
The models are available through the Meta Model API and inside Muse Code, Meta’s coding tool.
How they perform
On the independent Artificial Analysis index, Muse Wonder scores 52 at a cost of about 4.52 dollars per task. The larger Wonder Full scores 56 and ranks fourth among 689 models. For comparison, Claude Opus 5.5 scores 58.
Meta’s own tests show Wonder beating Opus 5.5 on long coding tasks. It falls behind on some other tests, such as Terminal-Bench and OSWorld. These are company-reported results, so independent testing will matter.
Pricing and access
Muse Wonder costs 4 dollars per million input tokens and 20 dollars per million output tokens. Anyone can try it through a self-serve API preview. Wonder Full has no public API and only runs inside Muse Code on the 50 and 200 dollar plans.
The launch is Meta’s clearest sign yet that its costly AI rebuild is producing real products. It comes in one of the busiest months for AI launches in recent memory.
Comments (0)
Be respectful — comments are moderated.