Meta launched two AI models on October 12: Muse Wonder and Muse Wonder Full. Wonder is available today through Meta's Model API and its Muse Code coding tool; Wonder Full is only in Muse Code, on its two top subscription plans. Meta's own pages say the Muse chat assistant will get Wonder "soon", with no date.

Who can use it today

Muse Wonder is on the Model API and on every Muse Code plan. Meta's docs say Muse Wonder Full is not on the API and needs a Power Usage or Maximum Usage plan. Setting reasoning effort to "max" on Wonder does not switch you to Full. Meta's models page labels Muse Wonder "in preview" on the API.

Both models take text, images, video and audio and return text, with a 1-million-token context window, per Meta's model card.

What it costs

On the API, Meta lists Muse Wonder at $4.00 per million input tokens, $20.00 per million output tokens, and $0.20 per million cached input tokens. Web search grounding is $2.50 per 1,000 queries. For comparison, Meta lists its earlier Muse Spark 1.3 at $1.25 input and $4.25 output.

Muse Code also offers pay-as-you-go billing. It has four flat-rate plans: Everyday, High, Power and Maximum Usage. Meta says Everyday allows 10 to 50 prompts every five hours, and the others offer 5x, 20x and 80x that usage. Meta's docs page does not list dollar prices; it says the plan details are shown during sign-up. We have not seen them.

The data trade-off

Muse Code subscriptions default to "Contributor mode": you get more usage, and Meta may use your prompts, code, files and the model's responses to train future Meta models. You can turn it off in settings. On the API, Wonder is Standard mode only, and Meta says content in Standard mode is not used to train its models.

Meta's launch post says Wonder has two agent features it calls "Active Memory" and "Active Focus", and says it was trained not to share private information without permission. Meta says its full safety report is "available soon".

How it compares, on Meta's numbers

Meta's table compares Wonder and Wonder Full against GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5, all at maximum reasoning effort except one SWE-Marathon entry for Wonder Full, which is marked xhigh. We did not test any of this, and some of the measures are Meta's own.

  • Artificial Analysis Intelligence Index: Wonder Full 56, Wonder 52, versus Opus 5.5 at 58 and Fable 5.1 and GPT-6 Astra at 53.
  • DeepSWE 1.1 (long coding tasks): Wonder Full 79.0, Wonder 78.8, versus 74.2 for Opus 5.5 and 74.1 for GPT-6 Astra.
  • ProgramBench (programming from scratch): Wonder Full 59.0, Wonder 50.0, versus 70.5 for Opus 5.5.
  • Terminal-Bench 4.0-AA: Wonder Full 57.1, versus 59.6 for Opus 5.5 and 59.1 for GPT-6 Astra.

So Meta's own table shows wins on some tests and losses on others. Meta's Artificial Analysis chart also lists Google's Gemini 4 Argon at 53, though Meta's table leaves it out, and The Verge says it is unclear how the models stack up against it.

What we could not confirm

We did not find Muse Code plan prices, a date for Wonder in the Muse assistant, or any statement that Wonder's weights will be released. We have not used either model.