ZeroDayBench: leaderboard
Metric: Pass rate (%) averaged over five information levels (zero-day, CWE hint, post-exploit report, one-day, full information); tasks: real high-severity CVEs ported into different open-source repositories; the agent (a bash and edit tool loop, up to 100 turns) must find and patch the vulnerability, and success means the patched code passes the exploit test; traces that ran git clone are excluded; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 56 | #138 |
| 2 | GPT-5.2 (Medium) | 48.2 | #105 (GPT-5.2) |
| 3 | Grok 4.1 Fast (Reasoning) | 34 | #208 (Grok 4.1 Fast) |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=zerodaybench · How It Works · Data refreshed daily, snapshot 2026-10-11.