AIMarketsNewsTech

Meta’s Muse Wonder launch puts frontier prices on its AI and keeps Wonder Full off the API

Meta on Monday, Oct. 12, released Muse Wonder and Muse Wonder Full, which it calls “our most capable models to date,” built for agents that carry out multi-step tasks such as working through an insurance claim or building an app, according to the company’s announcement. Developers can reach Muse Wonder through the Meta Model API at $4 per million input tokens and $20 per million output tokens, tokens being the chunks of words a model reads and writes, TechCrunch reported. Wonder Full “will not be served through the API.”

The only way to use Wonder Full is through Muse Code, Meta’s coding agent, on a $50-a-month “power usage” or $200-a-month “maximum usage” plan, Engadget reported. People using the Muse personal agent app will need a $20 or $100 monthly subscription to get the new model, according to CNBC.

TechCrunch calls it the company’s first attempt at a frontier model, the tier where OpenAI and Anthropic sell their best systems. Meta says Wonder Full beats Anthropic’s Claude Fable 5.1 and sits just below Claude Opus 5.5 on the Artificial Analysis Intelligence Index, a widely used combined score, Wired reported. Those rankings come from Meta’s own testing. Meta shares rose about 4% in afternoon trading, Reuters reported.

On price, the company and the press don’t quite agree. Meta AI chief Alexandr Wang told CNBC the API would be “meaningfully below” leading models from OpenAI and Anthropic. TechCrunch put the $4/$20 list price as roughly equivalent to GPT-6 Sol or Opus 5.5.

From download to subscription

Meta’s Llama models were, in Wired’s words, “the first frontier open-weight offerings from a big lab,” meaning anyone could download the model and run it on their own hardware. Researchers and startups adopted them widely. Meta still releases open-weight models, including the small Muse Glimmer, and says larger ones are coming. Wonder is a different product with a different business attached.

CNBC framed the launch as a test of whether users will pay, and as part of Meta’s push for revenue beyond ads. Meta has said it expects to spend as much as $145 billion this year on AI, per Engadget. Mark Zuckerberg told the Financial Times that reaching this point “required a large amount of compute and investment.”

Every developer gets Wonder by the token. Wonder Full, the model Meta is using to claim a spot near the top of the rankings, goes only to people who pay Meta a monthly fee to use Meta’s own coding tool.

An API customer buys raw model output and builds their own product on top, and the lab becomes a supplier that can be swapped out the day a cheaper model shows up. A Muse Code subscriber lives inside Meta’s interface, pays Meta every month, and has a reason to stay. Holding the best model back from the API means the outside builder who might compete with Muse Code can only rent the second-best version.

The safeguards follow the same line. CNBC reported that Meta’s strongest protections against cyber misuse come only with the paid model for now.

Not everything is in the open yet. Wired notes Meta did not disclose how the Wonder models were built or trained, and Meta says a full report with additional safety evaluations “will be available soon.”

The scale of the spending makes the subscription bet concrete. For scale, covering one year of Meta’s planned $145 billion at $200 a month would take about 60 million people on Muse Code’s top tier for the full year.

A few years ago a startup could take Meta’s best model, put it on its own servers and never send Meta a dollar. Today the best model Meta has comes with a login screen and a monthly bill from Meta.

Sources


Discover more from Bitter Fool

Subscribe to get the latest posts sent to your email.

Leave a Reply

Your email address will not be published. Required fields are marked *