AI from the other labs · 25 October 2026
Meta launches Muse Wonder, its first frontier-class models
Meta launched Muse Wonder and Muse Wonder Full on 12 October 2026, positioning them as its most capable models yet. Earlier checks on this topic missed the launch.
Both are aimed at agent work: coding, documents, and domain tasks in finance, law and health. Reports say they take text, images, audio and video, with a context window of about 1 million tokens (one outlet gives 1,048,576).
Who gets it. Muse Wonder is available to developers through Meta's Model API and Muse Code. Muse Wonder Full is limited to Muse Code subscribers. Meta says the models will later power parts of its Muse product, starting with subscribers.
What Meta highlights. Active Memory carries context across conversations. Active Focus helps a model keep going through multi-step tasks and recover from errors.
Unverified. One outlet lists API pricing for Muse Wonder at $4 per million input tokens and $20 per million output; I found no second source. Benchmark claims are Meta's own, and coverage notes they await independent testing. One report puts Artificial Analysis's index at 56 for Wonder Full and 52 for Wonder, below Claude Opus 5.5 at 58, but I could not corroborate that.
This is a different release from Muse Spark (April 2026, with a 1.1 update in July).
Sources