Bug Hunt Bench full leaderboard: DeepSeek V4.1 Flash fixes 42.7 bugs for under $2
PawelHuryn · x · 2026-11-10
Methodology FAQ and full leaderboard for Pawel Huryn's Bug Hunt Bench: real repos with re-planted bugs frontier models missed in early 2026, blind cross-family judges, only fully fixed planted bugs scored. Fable 5.2 max leads at 58/105 ($139.65), Sonnet 5.5 max 51.3 ($154.03). Standout value: DeepSeek V4.1 Flash 1017 max fixes 42.7 for just $1.84 (high tier: 39.2 for $1.27); GPT-6.1 Sol XHIGH 42.7 for $4.35. Meta's Muse Wonder Full 1.0 max scored 42; Kimi K3.1 Pro 36 ($13.76); Grok 4.7 29 ($22.96). Some OpenAI models report many irrelevant real issues to inflate findings — not counted. All run notes are open on GitHub.
Related event: Bug Hunt Bench: Fable 5.2 Tops Ranking, Fixing 58/105 Real Bugs(3 posts)→
More from coding & agent
- User has Codex sweep 191GB of disk garbage in under 3 minutes — justalexoki · 2026-11-10
- Harper: free open-source Grammarly alternative checks grammar offline in under 10ms — thisguyknowsai · 2026-11-10
- 1,000+ sign up for free project-based course on production AI agents for generative media — Al_Grigor · 2026-11-10
- The 12-item 2026 coding syllabus vs claims that coding is dead — JafarNajafov · 2026-11-10
- Dev's whirlwind week: OpenAI DevDay, Claude talk, hackathon runner-up — paw_lean · 2026-11-10
- Spacefast launches agent-ready platform with built-in databases, storage and domains — Scobleizer · 2026-11-10