Meta releases Muse Wonder: its first frontier models to power the Muse agent against Dots

#What happened

Meta's Superintelligence Lab (led by Alexandr Wang) released Muse Wonder and Muse Wonder Full on Monday, October 12, 2026, the company's first models it calls frontier. Two new capabilities headline: Active Memory, which carries context about a user across conversations and retrieves the relevant details when a task needs them, and Active Focus, meant to keep the model working when a task gets hard, try other approaches, recover from errors, and check in before following through on anything important. Both models offer a 1 million-token context window with text-only answers. On Meta's own testing they score comparably to Claude Fable 5.1 and GPT-6 Astra; independent evaluator Artificial Analysis scored Wonder Full 56 on its Intelligence Index (tying Claude Sonnet 5.5 and beating Fable 5.1, GPT-6 Astra, and Gemini 4 Argon, all at 53, but below Opus 5.5 at 58) and standard Wonder 52, one point behind Astra. Abacus.AI CEO Bindu Reddy publicly challenged the claims on X, saying Wonder was "nowhere near Fable or Astra level" with detailed results to follow. Neither model ships with downloadable weights, a break from Meta's Llama openness. Pricing: standard Wonder costs $4 per million input tokens and $20 per million output tokens through the Meta Model API (cached input discounted to $0.20 per million), roughly in line with GPT-6 Sol or Opus 5.5. Wonder Full is only available to Muse Code subscribers and not served through the API at all. In an FT interview timed to the launch, Mark Zuckerberg said "We will significantly accelerate the pace of model development." Meta's shares rose about 4% on the announcement.

#Why it matters

Wonder is built to power Meta's personal AI agent Muse, rolling out first to Muse subscribers, which puts Meta's best agent brain directly against OpenAI's Dots in the always-on personal-agent race. The closed, paid release also marks Meta's strategic shift: monetizing its top-tier models through subscriptions and API rather than open weights.

Verified October 15, 2026.