Meta's Muse Wonder release hits the AI frontier
Meta ships two frontier models at once: Wonder sells through the API at $4 per million input tokens, Wonder Full stays behind Muse Code.
In short
On October 12, 2026 Meta released its first frontier models, Muse Wonder and Muse Wonder Full, with Wonder served through the Meta Model API and Wonder Full reserved for Muse Code subscribers.
At a glance
- Released October 12, 2026: Muse Wonder and Muse Wonder Full, the first frontier models from Meta Superintelligence Labs.
- Meta Model API pricing: $4 per million input tokens and $20 per million output tokens.
- Wonder Full ships only to Muse Code subscribers, never through the API; it powers the Muse agent.
- Trained at Meta's Prometheus data center in central Ohio, which came online in 2026.
- MemoryBench is Meta's own benchmark; the wins over Fable, Opus and Astra come from Meta's own testing.
Meta released two frontier models on October 12, 2026. Muse Wonder sells through the Meta Model API; the stronger Muse Wonder Full never leaves the Muse Code subscription. They are the first models of this class to come out of Meta Superintelligence Labs.
What actually shipped
Wonder follows Muse Spark, a lightweight low-latency model, and the Muse agent. Wonder Full is what now drives that agent. In Meta's own measurements, as reported by TechCrunch, Wonder lands at scores comparable to Fable 5.1 and GPT-6 Astra.
Meta has published neither a parameter count nor a context window, and nothing in the coverage addresses whether the weights will be released. A break with the Llama approach is therefore neither documented nor ruled out.
The price, and the wall behind it
API access to Wonder costs $4 per million input tokens and $20 per million output tokens. Wonder Full is not served through the API at all. Reaching the stronger model means a Muse Code subscription, and Meta did not say what that subscription costs.
MemoryBench is Meta's own yardstick
The pitch centers on active memory. Meta introduced MemoryBench to score continual learning and memory construction, and says Wonder beats models from Fable, Opus and Astra on it.
Read that in order. The vendor wrote the benchmark and the vendor ran the test. No independent replication exists as of publication.
"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," says research scientist Kavya Kopparapu.
Prometheus, central Ohio
Training ran at Meta's Prometheus data center in central Ohio, which came online in 2026. The model generation rides on compute Meta added this year. The groundwork was the August 2025 reorganization of Meta's AI effort under Scale AI founder Alexandr Wang.
What we could not verify
Only one of three assigned reports loaded: TechCrunch. Two others were unreachable, so the benchmark results and the parity claim against Fable 5.1 and GPT-6 Astra stay single-sourced to Meta by way of that report. Parameter count, context window, the Muse Code price and the power draw at Prometheus are all undisclosed.
FAQ
How much does Muse Wonder cost?
$4 per million input tokens and $20 per million output tokens through the Meta Model API. Meta has not disclosed a price for the Muse Code subscription.
What is the difference between Muse Wonder and Muse Wonder Full?
Wonder is the API model. Wonder Full is the stronger one, ships only inside Muse Code, and powers the Muse agent.
Is Muse Wonder better than Fable 5.1 or GPT-6 Astra?
Meta reports comparable benchmark scores plus a lead on its own MemoryBench. None of that has been independently verified.
