UGI - Willingness (W/10): leaderboard
The Willingness column of the Uncensored General Intelligence leaderboard, rating how hard a model can be pressed on controversial prompts before it refuses or strays from the instructions. A direct component tracks outright refusals, and an adherence component tracks deviations, which the author notes can reflect either refusal or weak instruction following. Scored 0 to 10, higher means more willing.
Metric: W/10 Score. Source: huggingface.co. 1329 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 4B Instruct | 7.8 | #515 |
| 2 | Qwen2.5-32B-Instruct-abliterated-v2 | 7.8 | #394 |
| 3 | Mistral Large 2 (Nov) Instruct (2411) | 7.5 | #465 |
| 4 | Grok 4.20 0309 (Reasoning) | 7.5 | #112 (Grok 4.20 0309) |
| 5 | DeepSeek V3.2 (Non-reasoning) | 7.2 | #205 (DeepSeek V3.2) |
| 6 | Ministral-3-14B-Reasoning-2512 | 7.2 | #540 |
| 7 | jamba-large-1.7 | 7.2 | #525 |
| 8 | Mistral Medium 3 | 7.2 | #500 |
| 9 | DeepSeek V4 Flash (Non-reasoning) | 7.2 | #137 (DeepSeek V4 Flash) |
| 10 | GLM-4.5-Air (Non-reasoning) | 7 | |
| 11 | Qwen 3 4B (Non-reasoning) | 6.8 | #829 (Qwen 3 4B) |
| 12 | Devstral Small 2 | 6.8 | #495 |
| 13 | GLM-4.7 Flash (Non-reasoning) | 6.8 | #509 (GLM-4.7 Flash) |
| 14 | Gemmasutra-9B-v1 | 6.8 | #689 |
| 15 | Grok 3 | 6.5 | #304 |
Interactive version: theaggregate.ai/benchmark?slug=ugi-willingness-w-10 · How It Works · Data refreshed daily, snapshot 2026-10-20.