Muse Wonder is Meta's entry into the frontier tier, built for long agentic coding and tool-use sessions. It launched on the Meta Model API (still labelled preview in Meta's docs) at $4 input / $20 output per million tokens, the same price as Claude Opus 5.5, with a ~1M-token context window. Independent Artificial Analysis testing places it just behind Claude Opus 5.5 and Sonnet 5.5 overall. A larger Muse Wonder Full is only available inside Muse Code subscriptions, and Meta's separate Muse Spark Open release covers the open-weights side of the family.
Strong reasoning with extended thinking
1M token context window
Excellent code generation and debugging
Strong tool use and function calling
Reliable instruction following
Accepts image inputs
Strong creative writing ability
Broad multilingual support
| Benchmark | Score | Type | Recorded |
|---|---|---|---|
| Humanity's Last Exam | 50.1 | accuracy | today |
| SciCode | 60.3 | accuracy | today |
| LCR | 80.7 | accuracy | today |