CWE-bench v2 Held-Out Set Pass@1
Cybersecurity · 2026-11-02
Pass@1 on the 150 CWE-bench v2 tasks across 126 CWEs that Collinear AI keeps closed, averaging four rollouts per task.
Benchmark source: Cwe Bench
Top models (higher is better)
| Model | Score | Source |
|---|---|---|
| Claude Opus 5.5 | 67.0 | CWE-bench v2 leaderboard (snapshot 2026-11-03) · Collinear AI |
| Muse Wonder | 67.0 | CWE-bench v2 leaderboard (snapshot 2026-11-03) · Collinear AI |
| GPT-6 Astra | 64.5 | CWE-bench v2 leaderboard (snapshot 2026-11-03) · Collinear AI |
| Gemini 4 Argon | 63.2 | CWE-bench v2 leaderboard (snapshot 2026-11-03) · Collinear AI |
| DeepSeek V4.1 Flash 1017 | 58.0 | CWE-bench v2 leaderboard (snapshot 2026-11-03) · Collinear AI |
| Grok 4.7 | 54.2 | CWE-bench v2 leaderboard (snapshot 2026-11-03) · Collinear AI |
| Kimi K3.1 Pro | 47.8 | CWE-bench v2 leaderboard (snapshot 2026-11-03) · Collinear AI |
| Qwen3.8 Max (0902) | 45.7 | CWE-bench v2 leaderboard (snapshot 2026-11-03) · Collinear AI |
| Mistral Large 4 Preview | 45.2 | CWE-bench v2 leaderboard (snapshot 2026-11-03) · Collinear AI |
| Hy4 Preview | 44.2 | CWE-bench v2 leaderboard (snapshot 2026-11-03) · Collinear AI |
| GLM-5.3 | 40.7 | CWE-bench v2 leaderboard (snapshot 2026-11-03) · Collinear AI |