RefusalBench: leaderboard
Metric: Strict Refusal Rate (self-reported). Source: benchmarklist.com. 19 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.7 | 0.1 |
| 2 | GPT-4.1 Nano | 0.2 |
| 3 | Gemini 2.5 Pro | 0.4 |
| 4 | GPT-4.1 Mini | 0.6 |
| 5 | Gemini 2.5 Flash | 0.6 |
| 6 | Claude Opus 4.5 | 1.2 |
| 7 | Grok 4.20 | 1.3 |
| 8 | GPT-4.1 | 1.4 |
| 9 | GPT-4o | 2.1 |
| 10 | DeepSeek R1 | 18.4 |
| 11 | Qwen 2.5 Max | 29.8 |
| 12 | DeepSeek V3 (0324) | 47.5 |
| 13 | Mistral Large 2 (Jul) | 87.2 |
| 14 | Command A | 94.6 |
Interactive version: theaggregate.ai/benchmark?slug=refusalbench · How It Works · Data refreshed daily, snapshot 2026-09-05.