CWE-bench v1 - Language - Python - Programmatic Pass@1: leaderboard
Collinear CWE-bench archived v1, Language - Python - Programmatic 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: Programmatic Pass@1 (%). Source: cwe-bench.com. Saturation forecast: Around October 2027. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra | 71 | #6 |
| 2 | Grok 4.7 | 68 | #43 |
| 3 | Gemini 4 Carbon | 68 | #2 |
| 4 | Gemini 4 Argon | 57 | #11 |
| 5 | Grok 4.6 | 57 | #44 |
| 6 | GPT-6 Sol | 55 | #24 |
| 7 | Claude Fable 5.1 | 52 | #14 |
| 8 | Hy4 preview | 52 | #66 |
| 9 | Muse Spark 1.3 | 52 | #42 |
| 10 | Claude Opus 5.5 | 50 | #5 |
| 11 | DeepSeek V4.1 Flash | 46 | #58 |
| 12 | Inkling | 36 | #270 |
No result here: #1 Claude Fable 5.2, #10 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=cwe-bench-v1-language-python-programmatic-pass-1 · How It Works · Data refreshed daily, snapshot 2026-11-12.