ODCV-Bench - Incentivized: leaderboard
Same 40 scenarios with only the KPI pressure stated, not demanded; the number is the share of runs judged metric gaming or worse, so violations here are the agent's own initiative.
Metric: Misalignment rate (%, lower is better). Source: github.com. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 | 0 | #79 |
| 2 | Claude Opus 4.5 | 2.56 | #156 |
| 3 | GPT-5.1 Instant | 7.5 | #440 |
| 4 | Llama 4 Maverick | 15.79 | #1099 |
| 5 | GPT-OSS-20B | 22.5 | #1147 |
| 6 | GPT-5.5 | 25 | #49 |
| 7 | Kimi K2 0905 | 25 | #558 |
| 8 | GPT-OSS-120B | 27.5 | #700 |
| 9 | GLM-5.1 | 27.5 | #179 |
| 10 | Qwen 3.6 Max Preview | 27.5 | #159 |
| 11 | Qwen 3 Max | 28.21 | #403 |
| 12 | Kimi K2.6 | 30 | #178 |
| 13 | Qwen 3 30B A3B 2507 Instruct | 30.77 | #1089 |
| 14 | MiniMax-M2.7 | 30.77 | #518 |
| 15 | MiniMax-M2 | 35 | #674 |
Interactive version: theaggregate.ai/benchmark?slug=odcv-bench-incentivized · How It Works · Data refreshed daily, snapshot 2026-11-13.