Today · Tuesday, October 13, 2026
ModelsMetaOctober 12, 20262 min
Meta releases Muse Wonder for agents that remember and keep going
Company claim
Reported by Meta; benchmark comparisons are Meta's and are not independently verified.
Meta · October 12, 2026 · Primary
What happened
On October 12, 2026, Meta released Muse Wonder and Muse Wonder Full. Muse Wonder is available through the Meta Model API and Muse Code at $4 per million input tokens, $0.20 per million cached input tokens, and $20 per million output tokens. Muse Wonder Full is limited to Muse Code subscribers on the Power Usage and Maximum Usage plans. The Maximum plan is 80 times the Everyday Usage limit. The model card lists a 1 million token context, with text, image, video, and audio in, and text out. Meta says the Muse app will get Muse Wonder later, subscribers first. The models were trained at Meta's Prometheus data center.
Why it matters
This is the model Meta wants running agents that remember a person and stay on a long task, including coding. The part a developer can use today is the API price and Muse Code. The consumer app is still on the previous model until that later rollout. Meta says the pair lead on agent, memory, and coding tests. Those comparisons are Meta's. A fuller safety report is promised, not published with this post.
Go deeper
Meta describes two habits it wants from the model. Active Memory is carrying context about a person across conversations and pulling back the detail that matters. Active Focus is staying on a hard goal, trying another approach, and checking in before an important action.
The model card compares Muse Wonder and Muse Wonder Full with GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5. Meta says some of those competitor scores are its own runs and some are the other lab's published numbers. A separate safety report is still to come.
Sources
Reported by Meta; benchmark comparisons are Meta's and are not independently verified.
- Meta Primary · October 12, 2026