BioTIER Refusal
Safety · 2026-04-14
BioTIER Refusal measures refusal behavior on the BioTIER v1 refusal evaluation: 501 prompts spanning select agents and toxins, biomedical dual-use research of concern, and catastrophe-avoidance biological-risk topics. Scores are refusal rates in percentage points, so higher is safer.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5 | 100.0 |
| Sonnet 4.5 | 98.5 |
| Muse Spark 1.1 | 97.5 |
| Opus 4.7 | 94.7 |
| Opus 4.8 | 94.7 |
| Sonnet 4.6 | 94.2 |
| Opus 4.6 | 93.0 |
| Opus 4.5 | 91.7 |
| GPT-5.4 | 79.7 |
| GPT-5 | 78.6 |
| GPT-5 Nano | 78.6 |
| o3 | 75.6 |
| GPT-5.5 | 73.4 |
| Gemini 3.1 Pro Preview | 72.1 |
| GPT-5.6 Sol | 71.9 |
| Grok 4.20 | 71.6 |
| GPT-5.4 Mini | 71.4 |
| GPT-5.2 | 71.1 |
| O4 Mini | 71.1 |
| GPT-5.1 | 69.6 |
| GPT-5.6 Luna | 68.6 |
| Gemini 3 Flash Preview | 68.1 |
| GPT-5.6 Terra | 65.6 |
| GPT-5 Mini | 64.8 |
| GPT-5.4 Nano | 64.6 |
| Grok 4.1 Fast | 64.6 |
| O1 | 64.6 |
| Grok 4 Fast | 63.8 |
| Grok 4.3 | 63.6 |
| Grok 4 | 62.6 |
| Grok 4.5 | 62.6 |
| Gemini 3.5 Flash | 59.0 |
| Haiku 4.5 | 45.2 |
| GLM-5.2 | 42.2 |
| Kimi K3 | 38.4 |
| o3-mini | 35.2 |
| GLM-5.1 | 31.4 |
| Gemini 3.1 Flash-Lite Preview | 30.6 |
| Kimi K2.6 | 30.6 |
| GPT-4o (2024-11-20) | 21.9 |