Benchmarks · safety
DeepsecBench-AA Refusal Rate
Share of DeepsecBench-AA tasks the model refused on safety grounds. Lower means fewer refusals on defensive security work.
22 models · 23 results · data as of 2026-11-06 · artificialanalysis.ai ↗
Leaderboard
Fable 5.1 max
All results
23 rows| # | Model | Effort | Score |
|---|---|---|---|
| 1 | Fable 5.1 | max | 0.0% |
| 2 | Opus 5.5 | max | 0.0% |
| 3 | Opus 5 | max | 0.0% |
| 4 | DeepSeek V4.1 Flash 1017 | max | 0.0% |
| 5 | DeepSeek V4.1 Flash | max | 0.0% |
| 6 | Gemini 3.8 Flash | high | 0.0% |
| 7 | GLM 5.3 Flash | max | 0.0% |
| 8 | GLM 5.3 | max | 0.0% |
| 9 | GPT-6 Astra | max | 0.0% |
| 10 | GPT-6 Luna | max | 0.0% |
| 11 | GPT-6 Sol | max | 0.0% |
| 12 | Grok 4.7 | xhigh | 0.0% |
| 13 | Kimi K3.1 | max | 0.0% |
| 14 | Kimi K3 | max | 0.0% |
| 15 | MiMo V2.6 Pro | 0.0% | |
| 16 | MiniMax M3 | 0.0% | |
| 17 | Mistral Large 4 | high | 0.0% |
| 18 | Mistral Large 4 | medium | 0.0% |
| 19 | Muse Spark 1.3 | xhigh | 0.0% |
| 20 | Nemotron 3 Ultra | max | 0.0% |
| 21 | GPT-5.6 Terra | max | 9.1% |
| 22 | Qwen 3.8 27B | xhigh | 9.1% |
| 23 | Qwen 3.8 2.4T-A95B | 24.2% |