Meta releases its first frontier model, Muse Wonder

image via TechCrunch

October 12, 2026, 5:56 PM

• •Meta released two new models, Muse Wonder and Muse Wonder Full, as its first stated frontier-model attempt.
• •The company says the models deliver stronger benchmark performance than Muse Spark and improved results on complex coding tasks.
• •Meta highlighted “active memory” and cited MemoryBench results that it says exceed those of Fable, Opus, and Astra.
• •Muse Wonder costs $4 per million input tokens and $20 per million output tokens through the Meta Model API.
• •Muse Wonder Full will not be offered through the API and is reserved for Muse Code subscribers.
• •Meta says the models were trained at its Prometheus data center in central Ohio.

Meta’s Superintelligence Lab has introduced Muse Wonder and Muse Wonder Full, which it describes as its first frontier-model effort. The company says the larger models outperform its earlier Muse Spark system on complex tasks such as coding and post benchmark results comparable to rival top-tier models. Meta is emphasizing an “active memory” capability designed to help the models retain and synthesize user information over days, weeks, and months. Wonder is priced at $4 per million input tokens and $20 per million output tokens through the Meta Model API, while Wonder Full is limited to Muse Code subscribers and will also help power the Muse agent.

Read original article