Meta opens its first frontier model, Muse Wonder, via API

On October 12 Meta released Muse Wonder through its model API and OpenRouter. The stronger Muse Wonder Full is not on the API and is available only in Muse Code subscriptions starting at $50 a month. Meta claims both match the leading frontier models.

Muse Wonder costs $4 per million input tokens and $20 per million output tokens, with cached input at $0.20. The context window is 1,048,576 tokens, with no long-context premium. It accepts text, images, video, audio and PDFs and returns text.

Meta says the family is on par with GPT-6 Astra, Claude Opus 5.5 and Fable 5.1, but no independent benchmarks were published at launch, and Meta's research blog carried no technical post about the model on day one.

The context window is how much text a model can see in one request, enough here for long contracts or a whole codebase. Cached input lets a model reuse text it has already processed, sharply cutting costs for apps with fixed instructions or documents.

Meta was known for open-weight Llama models. Muse Wonder marks a shift to selling its best model through a paid API and subscriptions. A strong new provider adds price pressure, and free long context could push rivals' prices down for document-heavy work.

Since 2023 Meta released Llama as open weights that anyone could download and run, which built developer goodwill but little direct revenue. Over the past year it restructured its AI teams and paid record sums for top researchers to build frontier models. Muse Wonder is the first commercial product of that push, moving Meta closer to the paid model used by OpenAI and Google.

Revisit long-document tasks such as contract and report review, test the model on your own files rather than trusting launch figures, use caching where instructions repeat, and keep your setup provider-neutral so switching models stays a small change.