Meta launches Muse Wonder and Muse Wonder Full models for agents

Meta introduced two new AI models, Muse Wonder and Muse Wonder Full, on October 12, 2026. Muse Wonder is available through the Meta Model API and Muse Code, and Muse Wonder Full is available to Muse Code subscribers. Meta calls them "our most capable models to date" and says they are built for personal and professional agentic tasks, according to Meta's announcement of Muse Wonder. TechCrunch reported that the models are the company's "first attempt at a frontier model."

#What changed?

Meta added a new model series, Muse Wonder, with two models: Muse Wonder and Muse Wonder Full. The company says the models "deliver state-of-the-art performance on personal and professional agentic tasks." That is Meta's claim about its own product. The announcement itself gives no benchmark figures. TechCrunch reported that, "According to the company’s testing, Muse Wonder now scores higher on MemoryBench than Fable, Opus, or Astra." That result comes from Meta's own testing, not an independent evaluation.

Meta gives three example tasks: navigating an insurance claim, making sense of a lease and building a working app. It says the models bring that capability "to expert domains including law, finance, and health, along with frontier coding performance for people building everything from a game to an app for a new business."

The announcement names two features:

• Active Memory. Meta describes it as "the ability to carry what matters to you across conversations and bring back the details that are relevant to the task at hand." In a video accompanying the launch, research scientist Kavya Kopparapu said: "We put a tremendous amount of effort into making sure your model can remember what’s important to you over days, weeks, months, and beyond," according to TechCrunch.
• Active Focus. Meta describes it as the ability "to keep working toward your goal when a task gets hard, try other approaches, recover from errors, and follow through," and says the models ask for clarification when needed and check in before following through on anything important.

Meta says the new series is "built on the groundwork we laid with Muse Spark," its earlier model. The company ties the launch to an internal overhaul: "Fourteen months ago, we began an overhaul of our AI efforts, and in that time we’ve gone from rebuilding our foundations to pushing the frontier." TechCrunch reported, citing Meta's post, that the models were trained at Meta's Prometheus data center in central Ohio.

#Who can use it and when?

Both models were available on October 12, 2026, through different channels, according to Meta.

Meta says Muse Wonder "will soon power experiences in Muse, and will roll out first to subscribers." The announcement gives no date for that rollout.

TechCrunch reported that "Both models will also be used to power the Muse agent." Meta's own post names only Muse Wonder for Muse, so whether Muse Wonder Full will power Muse is not confirmed by Meta.

#How much does it cost?

TechCrunch reported that, "Accessed through the Meta Model API, Wonder will cost $4 per million input tokens and $20 per million output tokens." Meta's announcement does not list these prices. No price was reported for Muse Wonder Full or for a Muse Code subscription, and neither source gives regional availability.

#How to get it

Meta's announcement names the access routes but does not give setup steps.

1. Developers and builders can use Muse Wonder through the Meta Model API.
2. Muse Wonder is also available in Muse Code.
3. Muse Wonder Full requires a Muse Code subscription, per Meta. TechCrunch reported it is not available through the API.

For build and test details, Meta points readers to its research blog.

#How is Meta handling safety?

Meta says it trained the Muse Wonder family with specific limits for agent use. Because a personal agent can work across messages, calendars and documents, Meta says it "trained the Muse Wonder family to avoid sharing private information without permission; turn down requests to assist in performing actually harmful acts; and work to improve security and strengthen defenses against cyber attacks."

The company says it releases its most capable models "in line with the standards set out in our Meta Superintelligence Scaling Framework." It also says: "A full report with additional safety evaluations will be available soon." That report was not part of the October 12 announcement.

#How does it compare?

Meta positions Muse Wonder as a successor built on Muse Spark and as a step toward what it calls "personal superintelligence for everyone." Meta's announcement gives no head-to-head numbers. TechCrunch reported two comparisons from Meta's testing. Citing the model card, it said "Wonder and Wonder Full achieve comparable benchmark scores to Fable 5.1 and GPT-6 Astra." It also reported that Muse Wonder scores higher on MemoryBench than Fable, Opus or Astra. Astra is the model family behind OpenAI Dots, covered in our guide to GPT-6 Astra.

The practical difference for builders sits in access. Muse Wonder reaches developers through an API as well as Muse Code. Muse Wonder Full is limited to Muse Code subscribers, with no API access reported. Consumers in the Muse app have to wait for the subscriber-first rollout. For Muse's existing plans and availability, see our comparison of OpenAI Dots and Meta Muse; more coverage sits in the Assistants & Chatbots hub. Teams evaluating agent models can test Muse Wonder through the API at the reported $4 and $20 per million token rates, while Muse Wonder Full requires a Muse Code subscription.

#What is still unclear?

Several facts that buyers and builders need are missing from Meta's announcement:

• Pricing. Meta has not published prices; the API rates come from TechCrunch. No Muse Code subscription price was reported.
• Differences between the two models. Meta does not explain how Muse Wonder Full differs from Muse Wonder.
• Benchmarks and limits. Meta's announcement gives no benchmark results or context window size; the comparisons with Fable 5.1, GPT-6 Astra and other models on benchmarks and MemoryBench come from Meta's own testing as reported by TechCrunch.
• Muse rollout. Meta names only Muse Wonder for Muse; TechCrunch reported both models will power the Muse agent.
• Rollout timing. Meta says Muse Wonder will "soon" power Muse, with no date.
• Safety report. Meta says a full report with additional safety evaluations "will be available soon."

Meta says more detail on how the models were built and tested is on its research blog, which this article has not yet reviewed. DotsBot will update this article when those details are confirmed.