Meta releases Muse Wonder and Muse Wonder Full, its first frontier-class models, aimed at AI agents

What Meta released

Meta announced Muse Wonder and Muse Wonder Full on 12 October 2026 and described them as “our most capable models to date”.¹ TechCrunch reported that the models come from Meta’s Superintelligence Lab and are the company’s first attempt at a frontier model.² Meta says the models build on Muse Spark, its earlier and smaller model.¹²

Who can use them

Muse Wonder is available to developers through the Meta Model API and Muse Code, Meta’s coding agent. Muse Wonder Full is available only to Muse Code subscribers.¹³ Meta says Muse Wonder will “soon” power its Muse personal agent, starting with subscribers, but gave no date.¹

TechCrunch reported API pricing of $4.00 per million input tokens and $20.00 per million output tokens for Muse Wonder.² Meta’s announcement does not list prices.¹ SiliconANGLE reported a context window of one million tokens, with text, image, audio and video input and text output.³

What Meta says the models do

Meta highlights two features: “Active Memory”, which carries details across conversations, and “Active Focus”, which it says helps the model keep working on hard tasks, recover from errors and check in before important actions.¹³ Meta says the models perform well on work in law, finance, health and coding.¹

Benchmark claims

All benchmark figures come from Meta. According to SiliconANGLE’s summary of Meta’s results, Muse Wonder Full scored 57.5% on SWE-Marathon against 46.3% for Claude Opus 5.5, while Claude Opus 5.5 led on ProgramBench, 70.5% to 59%.³ Meta said its runs of competitor models were best-effort and not controlled comparisons.³ TechCrunch reported that Meta’s model card puts Wonder and Wonder Full roughly level with Fable 5.1 and GPT-6 Astra.² No independent tests were reported.²³

Safety

Meta says the models were developed under its Superintelligence Scaling Framework and that a fuller safety report will be published later.¹ SiliconANGLE reported that Meta does not rule out Muse Wonder Full reaching its high-risk threshold for chemical and biological threats without mitigations, and rates it “moderate or lower” with safeguards in place.³

What is not known

Meta has not published the full safety report or a date for the Muse app rollout.¹ Neither Meta nor the reports we read mention open weights.¹²³

Disclosure: Anawords uses Claude, made by Anthropic, as one of its research tools. Anthropic has no say in our coverage.