Meta Enters the Frontier Race With Muse Wonder at $4 Per Million Tokens

By Amara Osei

Meta will charge $4.00 per million input tokens and $20.00 per million output tokens for Muse Wonder, its first frontier AI model, released Monday by the company's Superintelligence Lab alongside a larger sibling, Muse Wonder Full. According to Meta's model card, both models post benchmark scores comparable to Fable 5.1 and GPT-6 Astra.

The pricing puts Wonder roughly on par with GPT-6 Sol or Opus 5.5. That's a deliberate step up from Muse Spark, Meta's lightweight, low-latency model, which prioritized speed over raw capability. The Wonder models are larger, score higher on benchmarks and perform better on complex tasks like coding, Meta said.

"Together, they mark the next step on our scaling ladder and a major leap toward personal superintelligence," Meta's launch blog post reads.

What's the distribution split?

Access works differently for the two models. Wonder runs through the Meta Model API at the published token rates. Wonder Full skips the API entirely — it's available only to Muse Code subscribers. Both models will also power the Muse agent, Meta's task-running assistant.

Why 'active memory' is the headline feature

Meta placed heavy emphasis on what it calls "active memory": the models' ability to retain and synthesize information a user gives them over long stretches of time. That fits the company's stated goal of building personal assistants that work across extended tasks rather than single-shot queries.

To back the claim, Meta cited its testing on MemoryBench, a benchmark that challenges continual learning and memory construction frameworks. The company says Muse Wonder now scores higher on MemoryBench than Fable, Opus or Astra.

"We put a tremendous amount of effort into making sure your model can remember what's important to you over days, weeks, months and beyond," research scientist Kavya Kopparapu said in a video accompanying the launch.

Where was it trained — and who's behind it?

The models were trained at Meta's Prometheus data center in central Ohio, which came online earlier this year, according to the post.

Monday's release is the first frontier model to ship since Scale AI founder Alexandr Wang restructured Meta's AI efforts in August of last year. Since that reorganization, Meta had released the Muse Spark model and the Muse agent but held back from the state-of-the-art releases that define OpenAI and Anthropic.

Wonder changes that posture. With benchmark parity claimed against GPT-6 Astra and Fable 5.1, and memory performance Meta says beats all three rivals, the release puts Wang's restructured lab into direct competition with the frontier leaders it had spent a year avoiding.