DeepsecBench — leaderboard
Vercel benchmark for finding vulnerabilities in application code. Models inspect 50 entry-point files against a hidden golden set of 231 human-judged findings; the primary score is recall-weighted F2, with runs repeated three times and reported by median.
Metric: Recall-weighted F2 score (%). Source: vercel.com. Status: years away from saturation. 31 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.6 Sol (xHigh) | 35.58 |
| 2 | Claude Opus 5 (Max) | 32.57 |
| 3 | GPT-5.6 Luna (xHigh) | 26.79 |
| 4 | GPT-5.6 Sol (Medium) | 25.1 |
| 5 | GPT-5.5 (xHigh) | 21.2 |
| 6 | GPT-5.6 Terra (xHigh) | 19.21 |
| 7 | Kimi K3 (High) | 17.56 |
| 8 | Claude Sonnet 5 (Max) | 17.06 |
| 9 | DeepSeek V4 Flash (High) | 16.54 |
| 10 | Grok 4.5 (High) | 15.58 |
| 11 | GPT-5.5 (Medium) | 12.63 |
| 12 | GPT-5.6 Terra (Medium) | 12.12 |
| 13 | GPT-5.6 Luna (Medium) | 12.08 |
| 14 | Gemini 3.6 Flash (High) | 8.99 |
| 15 | Claude Opus 4.8 (Max) | 6.91 |
Interactive version: theaggregate.ai/benchmark?slug=deepsecbench · How It Works · Data refreshed daily, snapshot 2026-08-05.