SEC-bench Pro - Firefox: leaderboard
SEC-bench Pro tests AI agents on finding vulnerabilities and writing working proof-of-concept inputs for real open-source code, each instance run in a Docker harness and confirmed by a checker. This board covers Firefox, which grew from 80 to 104 instances by June 2026: percent solved, timeouts counted as failures. Rows are agent, model and effort runs; the site withholds newer Claude Opus and Mythos results. Higher is better.
Metric: Firefox instances solved, headline mode (% of 104 instances; timeouts count as unsolved). Source: sec-bench.github.io. Saturation forecast: Around May 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 (xHigh) | 44.23 | #33 (GPT-5.5) |
| 2 | Claude Opus 4.6 (Max) | 27.88 | #67 (Claude Opus 4.6) |
| 3 | GPT-5.4 (xHigh) | 25 | #83 (GPT-5.4) |
| 4 | Kimi K2.5 (High) | 2.88 | #152 (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-pro-firefox · How It Works · Data refreshed daily, snapshot 2026-11-05.