Muse Wonder 1.0 tested on 105 bugs frontier models missed in Q1
curious_mahesh · x · 2026-10-13
Pawel Huryn tested Muse Wonder 1.0 on real work: 2 repos, 105 bugs frontier models missed in Q1 2026, find-and-fix. Maxed-out scores: GPT-6 Astra 45.0, Opus 5.5 41.7, Muse Spark 1.3 32.2, Muse Wonder 1.0 31.7, Grok 4.7 (xhigh) 29.0. A replier notes picking hard tail bugs is what actually separates rankings. The author distrusts maxed benchmarks: in his multi-run tests Wonder 1.0 is similar to Spark 1.3, a bit slower, and 50% pricier.