Meta is charging Muse app users $20 or $100 per month to access Muse Wonder, framing the frontier model release as a test of AI-derived revenue beyond its core advertising business5. Meta AI chief Alexandr Wang said the company developed Wonder, codenamed Watermelon, to improve the capabilities of "personal and professional agents" so they don't stumble on real-world tasks like booking dinner reservations.
Meta's Superintelligence Lab released Muse Wonder and Muse Wonder Full on October 12, the company's first frontier-class models, claiming benchmark parity with GPT-6 Astra, Claude Opus 5.5, and Claude Fable 5.12,9.
Fourteen-month rebuild
The launch caps a ground-up overhaul Meta said it began roughly fourteen months ago, moving from "rebuilding our stack from first principles to pushing the frontier"7. Mark Zuckerberg restructured Meta's AI research by investing billions in the recently formed Meta Superintelligence Labs and hiring former Scale AI CEO Alexandr Wang to lead it6. Where the earlier Muse Spark was a lightweight, low-latency model, the Wonder family is larger and targets complex tasks such as coding. Meta also released a small open-weight model called Muse Glimmer alongside the Wonder pair.
Pricing and access
Wonder is available through the Meta Model API at $4.00 per million input tokens and $20.00 per million output tokens, roughly equivalent to GPT-6 Sol or Opus 5.5. Wonder Full, the larger variant, is restricted to Muse Code subscribers and will not be served through the API. The Muse app itself is free, with subscription plans available for additional usage.
Active Memory and agentic focus
Meta highlighted two capabilities it calls Active Memory and Active Focus, designed to help AI systems perform useful work over extended periods and to keep the model locked into a stated goal while preventing confusion8. The company said it placed "particular emphasis" on the models' ability to retain and synthesize user-provided information over long interactions. Both models will power the Muse agent, starting with subscribers; Meta said it will develop specialized models it "can bring to everyone using Muse," though it has not shared specifics or a timeline1.
Benchmark positioning and safety
Meta's internal testing, presented in the model card, shows Wonder and Wonder Full achieving comparable scores to Claude Fable 5.1 and GPT-6 Astra. The company also cited Claude Opus 5.5 as a peer. How the models compare to Google's Gemini 4 Argon remains unclear, as that model's capabilities are "still coming into focus," The Verge noted.
Meta said it has put significant effort into training Wonder to refuse dangerous outputs such as bioweapons recipes or hacking assistance, and that the model offers its strongest safeguards to date against cyber misuse. Zuckerberg said last month that the company had delayed the release of its Muse models to work on safety.
ANALYSIS Pricing Wonder at the same tier as GPT-6 Sol and Opus 5.5 while restricting the larger Wonder Full to Muse Code subscribers creates a two-track distribution: API customers get a competitive but not top-end model, while paying Muse users get the full capability set. The structure ties Meta's strongest model directly to its own subscription product rather than to the open API market.
The $20-and-$100 consumer subscription tiers for Muse access layer a second monetization channel on top of the API, making the Wonder release as much a pricing experiment as a capability milestone.
Meta described the pair as "a major leap toward personal superintelligence".