dfbench: leaderboard

Models look for vulnerabilities in codebases they have not seen; the score is the F1 of their findings (balancing known vulnerabilities found against false reports), from depthfirst's own runs. Unlike CyberGym, which asks for a proof-of-concept input for one known bug, the model has to find the bugs itself.

Metric: Detection F1 on held-out codebases (%). Source: fireworks.ai. Saturation forecast: Not forecast. 10 models tracked. Updated 16 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Claude Opus 5.5High51.5#5
2DeepSeek V4.1 FlashHigh44.5#62
3Grok 4.5High38.9#85
4Kimi K3Max37#56
5Grok 4.7xHigh34#47
6MiMo-V2.6-Flash33.5#155
7GPT-5.6 LunaxHigh32.2#122
8GPT-5.6 SolxHigh28.3#30
9MiMo-V2.6-Pro27.2#76
10GPT-6 SolxHigh24.8#26

No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #6 GPT-6 Astra.

Interactive version: theaggregate.ai/benchmark?slug=dfbench · How It Works · Data refreshed daily, snapshot 2026-11-16.