CWE-bench v1 - Judge Panel Pass@1: leaderboard
Collinear CWE-bench archived v1, Judge Panel Pass@1, reported as a percentage. 120 private audit-and-patch tasks; four rollouts per task at high reasoning, one hour each; timeouts graded on code at timeout. Native model-specific harnesses. Programmatic success requires exploit blocked and existing tests passing. Separate majority panel of three rubric-based agentic judges checks correctness, happy-path, generalization, integrity and soundness; any failed axis rejects a fix. Pass@1 and Pass@4 remain separate. This is archived Collinear v1, not the Artificial Analysis Stirrup two-hour implementation or CWE-bench v2. Source: https://cwe-bench.com/v1
Metric: Judge Panel Pass@1 (%). Source: cwe-bench.com. Saturation forecast: Around April 2027. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 4 Carbon | 71 | #2 |
| 2 | Claude Opus 5.5 | 67 | #5 |
| 3 | GPT-6 Astra | 66 | #6 |
| 4 | Grok 4.7 | 64 | #43 |
| 5 | Gemini 4 Argon | 62 | #11 |
| 6 | Claude Fable 5.1 | 60 | #14 |
| 7 | Grok 4.6 | 55 | #44 |
| 8 | DeepSeek V4.1 Flash | 53 | #58 |
| 9 | GPT-6 Sol | 52 | #24 |
| 10 | Muse Spark 1.3 | 51 | #42 |
| 11 | Hy4 preview | 50 | #66 |
| 12 | Inkling | 33 | #270 |
No result here: #1 Claude Fable 5.2, #10 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=cwe-bench-v1-judge-panel-pass-1 · How It Works · Data refreshed daily, snapshot 2026-11-12.