DeepsecBench
Code · 2026-07-27
DeepsecBench is Vercel's AI Gateway leaderboard for the open-source deepsec security harness, which points coding agents at a fixed codebase snapshot to find vulnerabilities across 50 entry-point files. This historical definition matches findings against 231 human-judged issues and reports 100*5PR/(4P+R), selecting the median of the latest three successful runs for each model and effective reasoning level. The later 232-issue oracle regrades the same historical responses, so this version is a fallback component of the shared DeepsecBench factor rather than independent evidence.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.6 Sol | 35.6 |
| Claude Opus 5 | 32.6 |
| GPT-5.6 Luna | 26.8 |
| GPT-5.5 | 21.2 |
| GPT-5.6 Terra | 19.2 |
| Kimi K3 | 17.6 |
| DeepSeek-V4-Flash-0731 | 16.5 |
| Qwen3.8-Max | 16.5 |
| Grok 4.5 | 16.5 |
| GLM-5.3 Flash | 13.0 |
| Gemini 3.6 Flash | 11.6 |
| GLM-5.2 | 10.5 |
| Opus 4.8 | 6.9 |
| Inkling | 6.3 |
| Haiku 4.5 | 4.7 |
| Gemini 3.5 Flash-Lite | 2.1 |