SecRepoBench (Standalone) - Secure-Pass@1: leaderboard
SecRepoBench asks models to complete a masked region of a function in 318 tasks from 27 C/C++ GitHub repositories covering 15 CWEs, then compiles the project, runs its unit tests and replays an OSS-Fuzz proof-of-concept exploit. Standalone models get five BM25-retrieved functions as context. Secure-pass@1 is the percent of completions that pass the tests and survive the exploit; higher is better.
Metric: Completions that pass the unit tests and the PoC security test, pass@1 (%). Source: secrepobench.github.io. Saturation forecast: Around December 2027. 29 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 (Medium) | 39.3 | #105 (GPT-5) |
| 2 | O3 (Medium) | 32.4 | #132 (O3) |
| 3 | Claude Sonnet 4.5 (Thinking) | 31.1 | #148 (Claude Sonnet 4.5) |
| 4 | Claude Sonnet 4 (Thinking) | 30.2 | #207 (Claude Sonnet 4) |
| 5 | Claude 3.7 Sonnet (Thinking) | 28 | #259 (Claude 3.7 Sonnet) |
| 6 | GPT-4.1 | 27.7 | #258 |
| 7 | Gemini 3 Pro | 27.4 | #82 |
| 8 | O4 Mini (Medium) | 24.5 | #185 (O4 Mini) |
| 9 | DeepSeek R1 | 23.9 | #262 |
| 10 | O1 (Medium) | 23.6 | #190 (O1) |
| 11 | DeepSeek V3 | 22.6 | #329 |
| 12 | GPT-4o (2024-11-20) | 22 | #392 |
| 13 | GPT-OSS-120B (Medium) | 21.4 | #346 (GPT-OSS-120B) |
| 14 | O3 Mini (Medium) | 21.4 | #282 (O3 Mini) |
| 15 | Claude 3.5 Sonnet | 20.1 | #362 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=secrepobench-standalone-secure-pass-1 · How It Works · Data refreshed daily, snapshot 2026-11-05.