Meta has released its most capable models yet, a flagship step up from its lighter Muse Spark line and a new rival to the assistants your buyers already lean on.
Meta's first frontier models arrive in two sizes
On 12 October 2026, Meta released its first frontier-class models, the Muse Wonder series, a clear step up from the lighter Muse Spark. There are two variants: Muse Wonder and the more capable Muse Wonder Full. Both are agentic models built to power Muse, Meta's AI assistant. The work sits under Meta's Superintelligence Lab, which the release frames as part of its "Meta Superintelligence Scaling Framework."
Availability splits by product. From day one, Muse Wonder is reachable through the Meta Model API and through Muse Code, Meta's coding product. Muse Wonder Full is available to Muse Code subscribers only and is not on the API. Inside the Muse assistant itself the models are still rolling out, reaching subscribers first, so the app experiences are a work in progress rather than live everywhere.
The model card puts the technical shape of the release on record. Both models carry a one million token context window at inference and accept text, image, video and audio as input, returning text. Two features get Meta's billing. Active Memory carries context about a user across conversations and pulls back what is relevant to the task at hand. Active Focus is built to stay on hard, multi-step goals, trying alternatives, working through blockers and checking in before important actions.
On price, TechCrunch reports Muse Wonder on the Meta Model API at $4 per million input tokens and $20 per million output tokens, roughly in line with GPT-6 Sol and Claude Opus 5.5. Treat that as reported pricing, since Meta's own pages fetched did not restate it as a figure. As for how capable the models are, any benchmark numbers here are Meta's own. On its scorecard, drawing on Artificial Analysis and other third parties, Meta places Muse Wonder Full at 56 and Muse Wonder at 52, near Claude Fable 5.1 and GPT-6 Astra around 53 and below Claude Opus 5.5 at 58. Those are vendor-reported, not independent results.
A new rival assistant, not a change to how you are found
For a professional-services firm, this is a competitor model release worth knowing about, and little more than that this week. There is no ranking move, no setting to change and nothing to switch on. A more capable Meta assistant matters to your firm the way the recent OpenAI and Anthropic releases do: it widens the field of capable assistants that buyers may use to research and shortlist providers, and that may eventually sit inside Meta's own surfaces.
The steady response is the same one that has held through a busy year of model launches. Buyers increasingly ask an assistant before they ask around, which shapes how AI describes your firm. Being named by one of those assistants is becoming its own path to earning recommendations from AI, separate from ranking in a list of links. Keep your public pages clear, accurate and easy to quote, so whichever model reads them, Meta's included, describes you correctly. For firms that want help being found and cited in AI answers, that groundwork outlasts any single launch.
Frequently asked questions
What are Meta's Muse Wonder and Muse Wonder Full models?
Muse Wonder and Muse Wonder Full are Meta's first frontier-class AI models, released on 12 October 2026 as a step up from its lighter Muse Spark line. Both are agentic models built to power Muse, Meta's AI assistant, and both carry a one million token context window and accept text, image, video and audio as input. Meta's Superintelligence Lab built them.
What is the difference between Muse Wonder and Muse Wonder Full?
Muse Wonder Full is the more capable of Meta's two frontier models, and Muse Wonder is the standard version. They also differ in where you can use them: from launch on 12 October 2026, Muse Wonder is available through the Meta Model API and Muse Code, while Muse Wonder Full is limited to Muse Code subscribers and is not offered on the API. On Meta's own vendor-reported scorecard, Muse Wonder Full scores a little higher than Muse Wonder.
How much does Meta's Muse Wonder cost, and where can you use it?
TechCrunch reports Muse Wonder on the Meta Model API at $4 per million input tokens and $20 per million output tokens, which Meta's own pages did not restate as a figure, so treat it as reported pricing. On release day Muse Wonder runs through the Meta Model API and Muse Code, Muse Wonder Full through Muse Code subscriptions only, and both are still rolling out inside the Muse assistant, subscribers first.
How does Meta's Muse Wonder compare with other frontier AI models?
Meta positions Muse Wonder as competitive with the current flagships, but the comparison rests on its own figures. On Meta's scorecard, which draws on Artificial Analysis and other third parties, Muse Wonder Full scores 56 and Muse Wonder 52 on the Artificial Analysis Intelligence Index, near Claude Fable 5.1 and GPT-6 Astra at around 53 and below Claude Opus 5.5 at 58. These are vendor-reported numbers from Meta, not independent benchmarks.
What does Meta's Muse Wonder release mean for firms that want to be found?
Meta's Muse Wonder release does not change how your firm ranks or how it gets found, and there is nothing to action. It adds another capable assistant to the field buyers may use when they research and shortlist providers, and one that may later sit inside Meta's surfaces. The useful move is unchanged: keep your public pages clear, accurate and easy to quote so any assistant reading them describes your firm correctly.
