First independent tests of Muse Wonder: Wonder Full scores 56 and ties Sonnet 5.5 on coding agents, but at $4.52 per task it costs more than GPT-6 Astra and Gemini 4 Argon

Why it matters

【Key points】

About an hour after Meta's launch, Artificial Analysis published the first independent results: on its Intelligence Index v4.3, Muse Wonder (max) scores 52 and Muse Wonder Full (max) 56, versus 48 for Muse Spark 1.3 (max), released only six weeks earlier. Vals AI says Muse Wonder Full (Max) debuts at #5 on the Vals Index, up 6.4 points, ahead of every OpenAI model.

The gains are concentrated in agentic work. Wonder Full beats its predecessor on all ten index evals: Terminal-Bench 4.0 rises from 33% to 57%, GDPval-AA from 1681 to 1857 Elo (#2), AA-Briefcase to 1729 Elo (#3), and it tops CritPt at 33%. On the Coding Agents Index, Wonder Full in Muse Code scores 68, level with Claude Code on Sonnet 5.5 (68); Opus 5.5 scores 66.

It is expensive. Muse Wonder costs $4 per million input tokens, $0.20 cached and $20 output. With 7.69M input tokens per index task, that is $4.52 per task: about 2.8x its predecessor ($1.60), above GPT-6 Astra ($3.26) and Gemini 4 Argon ($1.99), below Claude Opus 5.5 ($5.98). Wonder Full is only in Muse Code with no per-token price, so its cost per task can't be computed yet.

【Verifiable】

Artificial Analysis: https://x.com/ArtificialAnlys/status/2109697083519447118

Vals AI: https://x.com/ValsAI/status/2109706320383545474

Meta announcement: https://about.fb.com/news/2026/10/introducing-muse-wonder-models/

【Our read】

1. The independent numbers largely back Meta's pitch: the progress is in long, hands-on agent tasks rather than chat, matching its Active Focus framing.

2. The strongest model, Wonder Full, is only available inside Meta's own coding subscription. Developers can't buy it by API yet, so the headline scores and the purchasable product are not the same thing.

3. Meta is not starting a price war; it is charging in Anthropic's bracket. Whether it wins developers depends on Full's API price, the promised safety report, and whether other leaderboards reproduce these rankings.