ODCV-Bench - Severity: leaderboard
Same runs as ODCV-Bench; the number is the mean judged severity on the 0-5 scale (0 legitimate, 3 metric gaming, 4 data falsification, 5 catastrophic) rather than the share over the threshold.
Metric: Mean judged severity, 0-5 (lower is better). Source: github.com. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 | 0.01 | #79 |
| 2 | Claude Opus 4.5 | 0.07 | #156 |
| 3 | GPT-5.1 Instant | 0.39 | #440 |
| 4 | GPT-5.5 | 0.71 | #49 |
| 5 | GLM-5.1 | 0.92 | #179 |
| 6 | Qwen 3.6 Max Preview | 1.18 | #159 |
| 7 | Kimi K2.6 | 1.34 | #178 |
| 8 | MiniMax-M2.7 | 1.37 | #518 |
| 9 | Kimi K2 0905 | 1.47 | #558 |
| 10 | GPT-OSS-20B | 1.48 | #1147 |
| 11 | Qwen 3 Max | 1.49 | #403 |
| 12 | Grok 4.1 Fast | 1.56 | #430 |
| 13 | GPT-OSS-120B | 1.57 | #700 |
| 14 | Qwen 3.6 27B | 1.67 | #341 |
| 15 | MiniMax-M2 | 1.68 | #674 |
Interactive version: theaggregate.ai/benchmark?slug=odcv-bench-severity · How It Works · Data refreshed daily, snapshot 2026-11-13.