Muse Wonder vs Claude Opus 5.5 vs Fable 5.1 vs GPT-6 Astra vs Gemini 4 Argon

Muse Wonder is Meta's first frontier model from Meta Superintelligence Labs, built for agentic and coding work. It takes text, image, audio, video and PDF input, with a 1M-token context.

Same price as Opus 5.5: $4 / $20 per 1M tokens, $0.20 cached input, and web search at $2.50 per 1,000 queries.

Two models, one name. The Meta Model API serves Muse Wonder. The stronger Muse Wonder Full, which scores higher in Meta's headline benchmarks, is available only to Muse Code subscribers.

Independent view (Artificial Analysis Intelligence Index):

• Opus 5.5 (max) 57.6 · Wonder Full 55.8 · Fable 5.1 53.4 · GPT-6 Astra 52.7 · Gemini 4 Argon 52.6 · Muse Wonder 52.4
• Speed is its clearest win: about 161 tokens/s median output, against 95 for Opus 5.5, 85 for Argon, 69 for Fable 5.1 and 66 for Astra
• Cost per index task: $4.52, below Opus 5.5 ($5.98) and Fable 5.1 ($7.63) but above Astra ($3.26) and Argon ($1.99)

From Meta's model card (self-reported, 14 benchmarks): Wonder Full has the top score on 6, Opus 5.5 on 6, Muse Wonder on 1 (MemoryBench) and Astra on 1. Coding is the API model's strong spot: on DeepSWE 1.1 Muse Wonder scores 78.8, close to Wonder Full (79.0) and ahead of Opus 5.5 (74.2) and Astra (74.1). Opus 5.5 leads Terminal-Bench 4.0 (59.6 vs 52.0) and ProgramBench (70.5 vs 50.0).

Bottom line: the API model costs as much as Opus 5.5. Its overall index score sits closer to Astra and Argon, but it is strong on long-horizon coding and fast.

Introducing Muse Wonder (Meta) · Model card · Meta Model API