Open-Source 763B Model Tested on 105 Hidden Bugs

Pawel Huryn's Bug Hunt Bench planted 105 hidden bugs across two real codebases to test frontier models' debugging ability. Muse Wonder Full 1.0 led with 42/105, Opus 5.5 Max scored 41, and the open-source 763B DeepSeek V4.1 Flash fixed 34 while approaching Opus at roughly 1/150 of the cost.

2026-10-17 ~ 2026-10-17 · 3 related posts

• DeepSeek V4.1 Flash 1017 fixes 34/105 hidden bugs in real-repo test, weights on Hugging Face — PawelHuryn · 2026-10-17
• 105 real bugs benchmarked: Opus 5.5 and Muse Wonder lead, DeepSeek V4.1 Flash 1017 jumps to 34 — PawelHuryn · 2026-10-17
• DeepSeek V4.1 Flash 1017 nears Opus 5.5 on bug-hunting bench at 1/150th the cost — PawelHuryn · 2026-10-17