Meta ships Muse Wonder frontier models at $4 per million tokens

Meta Superintelligence Labs released Muse Wonder and Muse Wonder Full on Monday, its first models that Meta calls frontier-class, priced at $4 per million input tokens and $20 per million output tokens on the Meta Model API. The larger Muse Wonder Full is not sold through the API at all and runs only inside Muse Code, Meta's coding tool, on subscription plans at $50 and $200 a month.

Meta said in its newsroom post that Muse Wonder is available today for developers and builders, while Muse Wonder Full is available today for Muse Code subscribers. Meta's developer page lists a one-million-token context window for both models and cached input at $0.20 per million tokens, with no API rate published for Muse Wonder Full.

Chief executive Mark Zuckerberg announced the release on X, writing that Muse Spark was Meta's first family of models and that Muse Wonder is "a much larger and more advanced class of models that brings us to the frontier." Meta's research blog says the models were trained at its Prometheus data center, which TechCrunch places in central Ohio, and carry two features it labels Active Memory and Active Focus.

Meta's price sits above its own last release and beside the models it names as rivals. VentureBeat's price table lists Muse Spark 1.3 at $1.25 per million input tokens and $4.25 per million output, Claude Opus 5.5 at the same $4/$20 as Muse Wonder, and OpenAI's GPT-6 Astra and Anthropic's Claude Fable 5.1 at $10/$50 each. TechCrunch described Wonder's rate as making it "roughly equivalent in cost to GPT-6 Sol or Opus 5.5." AI Tools Recap framed the same $4/$20 rate as double the $2/$10 it reports for GPT-6.1 Sol and Claude Sonnet 5.5, a harder reading of the rate than TechCrunch's "roughly equivalent" description.

Access marks the bigger change. Neither Wonder model ships with downloadable weights, breaking from the Llama series Meta released for anyone to run, and Muse Wonder Full has no pay-as-you-go rate and cannot yet be deployed outside Muse Code, VentureBeat reported. Meta said a full report with additional safety evaluations will be available soon.

The performance numbers are partly independent. Meta published a benchmark table for both models, and the evaluator Artificial Analysis scored Muse Wonder Full 56 and Muse Wonder 52 on its Intelligence Index, VentureBeat reported. That 56 ties Anthropic's Claude Sonnet 5.5 and sits behind Claude Opus 5.5 at 58. Meta did not disclose a parameter count for either model.

Consumer access is cheaper than developer access. CNBC reported that Muse app users pay $20 or $100 a month, while Muse Wonder Full requires Muse Code's $50 Power Usage or $200 Maximum Usage tier. Alexandr Wang, who leads Meta Superintelligence Labs, told CNBC the API offering is a "much larger model and at frontier level" and will stay priced "meaningfully below" leading models from OpenAI and Anthropic, a claim that sits awkwardly beside the $4/$20 rate TechCrunch calls equivalent to Opus 5.5.

For creators, the pitch is not a discount. A video or marketing team already paying a rival by the task has no cost reason to switch, because $4/$20 is not cheaper than Claude Opus 5.5. If the benchmarks hold up, the draw is memory and long-running agent work rather than price, and Muse Wonder Full is the version creators cannot reach through an API at all.

Meta says a full report with additional safety evaluations will follow, and that Muse Wonder will reach the Muse app for subscribers first, without naming a date.