SYLIN

Muse Wonder

by Meta · Model · Language & code models

mixed

Fast and honest about what it doesn't know, but costly and behind the leaders

Meta's frontier model for agentic and professional work; Muse Wonder Full for Muse Code subscribers, only Muse Wonder via API

AccessAPI

Why we say so

Meta's own benchmarks claim wins over GPT Astra and Claude Opus in some agentic and knowledge-work tests. Artificial Analysis found it fast but expensive and less intelligent than GPT-6 Astra or Gemini 4, though it hallucinates far less. LiveBench puts it outside the top 10, while Vending Bench ranks it fourth and Vals Index fifth.

Evidence

  1. Oct 18
    Meta's self-reported benchmarks show it beating top GPT and Claude models in some knowledge-work and agentic-coding tests
  2. Oct 18
    Fast but expensive, around 450 per task on Artificial Analysis, and less intelligent than GPT-6 Astra or Gemini 4

Alternatives

Gemini NeonTops a secure-code benchmark, just ahead of Gemini 4 Carbon, then the two Claude models, GPT-6 Astra and Muse Wonder
Made from