Muse Wonder disappoints on RareBench: Opus 5.5 leads at 52% recall

danielmckinn0n · x · 2026-10-13

GamowLabs' RareBench (122 cases) results for the newly released Muse Wonder checkpoint are underwhelming:

• Muse Wonder High scores 19% recall, a small regression vs Muse Spark 1.3 Max (22%), and far behind frontier models (no replicates, so not statistically significant)
• Leaders: Opus 5.5 at 52%, Sonnet 5.5 40%, Fable 5.1 46%, GPT-6 Astral 38%
• The author argues RareBench is a verifiable, long-horizon eval requiring dozens of diverse tool calls, harder to game than public indices
• This was the weaker checkpoint; the stronger Muse Wonder Full will be benchmarked once API access is granted

A bearish signal for the Muse line, unless Meta is simply optimizing for different tasks.

Original post →

More from Models

• Anthropic says new defenses cut distillation-harvested data by 90-99%, names Zhipu's GLM-5.3 — scaling01 · 2026-10-13
• Anthropic says it cut distiller data harvesting by 90-99%, turning away trillions of tokens — abhiadesai · 2026-10-13
• Ex-exMVP: post-training should add a facet for how well models explain things — MParakhin · 2026-10-13
• GLM-5.3 abliterated for a month, no takeover: dev rails against AI doom narratives — mariofilhoml · 2026-10-13
• Claude auto-files a bug report about itself: TLDRs still too long after user asked for shorter — andrey_kurenkov · 2026-10-13
• BridgeBench one-prompt test shows Muse Wonder is fast and cheap but worst-looking — PVORY · 2026-10-13

Models channel →