Meta wants back in the frontier AI conversation, and its latest models suggest the company is serious about it.
On October 12, 2026, Meta released Muse Wonder and Muse Wonder Full, two models built for personal and professional work. On the Artificial Analysis Intelligence Index v4.3, Muse Wonder Full scored 56 at maximum reasoning effort. That makes it the second highest-rated model on the index, behind only Anthropic’s Claude Opus 5.5.
How the Muse Wonder models stack up
The scores show a clear pecking order within Meta’s new lineup. Muse Wonder Full posted 56 on the index at max effort, while the standard Muse Wonder came in at 52.
The bigger number for Meta watchers is the jump from its previous model. Muse Spark 1.3 scored 48 on the same index, so Muse Wonder Full represents an 8-point improvement.
Muse Wonder Full also ranked in the top three on six of the ten evaluations that make up the Intelligence Index. Its strongest showings came in agentic tasks and coding benchmarks.
According to the post announcing the results, Muse Wonder outperforms Claude Opus 5.5 on most benchmarks. It also reportedly surpasses OpenAI’s GPT-6 Astra and Anthropic’s Claude Fable 5.1 on coding, legal, and automation tests.
On the aggregate Artificial Analysis score, though, Claude Opus 5.5 still holds the top spot.
Active Memory, Active Focus, and the price tag
Meta paired the models with two features aimed at a persistent problem for AI agents: reliability. They’re called Active Memory and Active Focus, and both are designed to make the models more dependable during agentic tasks.
Distribution is split by tier. Muse Wonder is available through the Meta Model API and Muse Code, while Muse Wonder Full is restricted to subscribers.
Pricing for Muse Wonder is set at $4 per million input tokens and $20 per million output tokens.
A comeback story built on a 2025 overhaul
The Muse Wonder release reflects how quickly Meta has accelerated its AI development since a 2025 overhaul of its efforts. The company restructured its approach that year, and these models are the most visible result so far.
Meta is positioning the Muse Wonder family as a direct challenge to the current leaders in AI, particularly OpenAI and Anthropic.
What this means for the AI race
The agentic focus is the part worth watching most closely. Meta’s strongest results came precisely in agentic tasks and coding, and features like Active Memory and Active Focus suggest the company is betting that dependability will be a key factor in that market.
The split between Muse Wonder and the subscriber-only Muse Wonder Full means the version that scored 56 is not the one every developer can access through the API.
At $4 per million input tokens and $20 per million output tokens for Muse Wonder, developers will compare cost against performance gains before switching providers.
For Meta, the jump from 48 to 56 in a single generation is the strongest evidence yet that its 2025 restructuring is producing results.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.