EvasionBench: leaderboard
Benchmark for evaluating model robustness against evasion-style safety and policy-circumvention prompts.
Metric: Macro-F1 (%). Source: iiiiqiiii.github.io. Status: saturation imminent. 12 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Flash | 84.64 |
| 2 | Claude Opus 4.5 | 84.38 |
| 3 | GLM-4.7 | 82.91 |
| 4 | GPT-5.2 | 80.9 |
| 5 | Qwen 3 Coder 480B A35B Instruct | 78.16 |
| 6 | Qwen3 Coder | 78.16 |
| 7 | MiniMax-M2.1 | 71.31 |
| 8 | DeepSeek V3.2 | 66.88 |
| 9 | Kimi K2 | 66.68 |
| 10 | Kimi K2 0905 | 66.68 |
Interactive version: theaggregate.ai/benchmark?slug=evasionbench · How It Works · Data refreshed daily, snapshot 2026-09-05.