Meta Launches Muse Wonder and Muse Wonder Full, Claims Frontier AI Status

What happened

On October 12, Meta unveiled two models, Muse Wonder and Muse Wonder Full. Meta said they are its most capable models yet, positioned to compete with the top offerings from OpenAI and Anthropic. They are the second model family from Meta Superintelligence Labs, following Muse Spark, which launched in April. The new models were reportedly code-named Watermelon before launch.

Zuckerberg wrote on social media that Muse Spark was the lab's first model family, and Muse Wonder is "a much larger and more advanced class of models that brings us to the frontier." In its official blog post, Meta said Wonder performs comparably to Claude Fable and GPT-6 Astra on legal, financial, and coding tasks.

Key facts

1. Availability: Muse Wonder is available to developers through the Meta Model API and Muse Code; Wonder Full is limited to Muse Code subscribers. Meta's Muse personal AI agent will get Wonder first for paying subscribers.
2. Pricing: via the API, Muse Wonder costs $4 per million input tokens and $20 per million output tokens.
3. Two features: Active Memory carries user context across conversations and recalls it when relevant; Active Focus keeps an agent working toward a goal through blockers and checks in before consequential actions.
4. Benchmarks: Meta says its model card shows results comparable to Claude Fable 5.1 and GPT-6 Astra; independent tester Artificial Analysis scored Wonder Full 56 on its Intelligence Index, 2 points behind Claude Opus 5.5.
5. Market reaction: Meta shares rose about 4% in afternoon trading on the day of the announcement.