SEC-bench: leaderboard
Security engineering benchmark for AI agents; its site now tracks SEC-bench Pro, where agents find vulnerabilities in V8, Firefox and Linux and write proof-of-concept inputs confirmed by a checker.
Metric: Instances solved, headline mode (% of 344 instances across V8, Firefox and Linux; timeouts count as unsolved). Source: sec-bench.github.io. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 (xHigh) | 58.43 | #31 (GPT-5.5) |
| 2 | GPT-5.4 (xHigh) | 38.95 | #79 (GPT-5.4) |
| 3 | Claude Opus 4.6 (Max) | 30.81 | #63 (Claude Opus 4.6) |
| 4 | Kimi K2.5 (High) | 2.33 | #144 (Kimi K2.5) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=sec-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.