Meta introduced the Muse Wonder series on October 12, 2026, describing Muse Wonder and Muse Wonder Full as its most capable models to date, built for personal and professional agentic tasks. The company said the models are built on the groundwork laid with Muse Spark.
Muse Wonder is available through the Meta Model API and Muse Code. Muse Wonder Full is available to Muse Code subscribers. Meta said Muse Wonder will soon power experiences in Muse and will roll out first to subscribers.
What the models are designed to do
Meta describes the models as targeting expert domains including law, finance and health, along with frontier coding performance. The examples the company gives are navigating an insurance claim, understanding a lease and making sense of someone’s finances.
Two capabilities are given specific names. Active Memory carries what matters to you across conversations and brings back details relevant to the current task. Active Focus is described as the ability to keep working toward a goal when a task gets hard, try other approaches, recover from errors, check in before following through on anything important and ask for clarification when needed.
How Meta describes its safety approach
Meta says it trained the Muse Wonder family to avoid sharing private information without permission, to turn down requests to assist in actually harmful acts, and to work toward improving security. The company also says developers and security teams can use models like Muse Wonder to find flaws in their own code, so it made sure the models can still do that work.
Meta says it trains, evaluates and releases its most capable models in line with the standards in its Meta Superintelligence Scaling Framework. A model card describing evaluations and the limits of that testing is published alongside the announcement.
What this means in practice
Two things are worth separating. Availability is real and specific: Muse Wonder through the API and Muse Code, Muse Wonder Full through a Muse Code subscription. The capability claims are Meta’s own descriptions of what the models are designed to do, not independent test results.
If you are choosing a model for agentic work, the practical question is whether the task needs memory across sessions and whether you want the model to keep going on a hard task without asking you. Those are the two features Meta is drawing attention to, and they are the ones worth testing on your own workload before committing.
For a related angle on how AI is being embedded in consumer hardware, our coverage of Meta’s open Muse Spark work looks at the local-AI side of the same strategy.
What we did not do
We did not benchmark the models, run the API, or verify the capability claims independently. All performance and safety statements come from Meta’s own announcement and model card. Pricing for the API and subscriptions was not part of the announcement we reviewed.
Sources
- Meta — Introducing Muse Wonder and Muse Wonder Full: Frontier Models for Agents
- Meta — Muse Wonder Model Card
- Meta — Put your ideas to work with Muse Wonder and Muse Wonder Full
- Honest Fred — OpenAI Acquires Orchestra Bio
