Benchmarks · safety
UGI
A private-question suite combining sensitive-topic knowledge with willingness to follow controversial instructions. Maintained by DontPlanToEnd. Score: UGI 🏆.
167 models · 168 results · data as of 2026-10-29 · huggingface.co ↗
Leaderboard
Grok 4.20 xhigh
All results
168 rows| # | Model | Effort | UGI |
|---|---|---|---|
| 1 | Grok 4.20 | xhigh | 70.0 |
| 2 | Grok 4.0709 | 67.8 | |
| 3 | Grok 3 | none | 63.5 |
| 4 | Grok 4.5 | max | 62.3 |
| 5 | Grok 4.3 | xhigh | 59.4 |
| 6 | Mistral Large 2411 | 59.2 | |
| 7 | DeepSeek V4 Flash | max | 59.2 |
| 8 | DeepSeek V3 | none | 57.0 |
| 9 | Mistral Large 3 | none | 56.9 |
| 10 | Kimi K2 | medium | 56.5 |
| 11 | Mistral Medium 3 | none | 56.5 |
| 12 | DeepSeek V3.0324 | 55.0 | |
| 13 | Opus 5.5 | max | 54.5 |
| 14 | DeepSeek V3.2 | high | 53.8 |
| 15 | Opus 4.7 | max | 53.6 |
| 16 | GPT-6.1 Sol | max | 53.1 |
| 17 | o3 | high | 52.1 |
| 18 | Mistral Large 2407 | 51.9 | |
| 19 | Sonnet 4.6 | max | 51.9 |
| 20 | Opus 4.6 | max | 51.6 |
| 21 | DeepSeek V3.1 | none | 51.3 |
| 22 | DeepSeek R1 | xhigh | 51.0 |
| 23 | Opus 4.8 | max | 50.7 |
| 24 | GPT-5.5 | xhigh | 50.0 |
| 25 | Mistral Medium 3.5 | high | 49.9 |
| 26 | Mistral Medium 3.5 | none | 49.5 |
| 27 | Llama 7 0b | 49.4 | |
| 28 | GLM 4.7 | none | 49.4 |
| 29 | Gemini 3 Flash | high | 48.9 |
| 30 | GPT-6 Astra | max | 48.8 |