UGI Leaderboard: leaderboard
A private question set tests what a model knows in three sensitive areas: hazardous subjects, adult or controversial entertainment, and socio-political topics. A separate willingness measure records how readily it answers instead of refusing or straying from the instructions. The published number combines the two, so it is an index over several tests rather than one task, and higher is better.
Metric: UGI Score. Source: huggingface.co. Saturation forecast: Around April 2028. 1329 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Grok 4.20 Multi-Agent | 70 | #113 |
| 2 | Grok 4 | 67.83 | #181 |
| 3 | DeepSeek V3.2 Speciale | 65.37 | #128 |
| 4 | Grok 4.20 0309 (Reasoning) | 64.23 | #112 (Grok 4.20 0309) |
| 5 | Grok 3 | 63.47 | #304 |
| 6 | Grok 4.1 Fast (Reasoning) | 63.24 | #217 (Grok 4.1 Fast) |
| 7 | Grok 4.5 | 62.32 | #48 |
| 8 | DeepSeek V4 Pro (Reasoning) | 62.26 | #102 (DeepSeek V4 Pro) |
| 9 | Grok 4.3 | 59.41 | #188 |
| 10 | Mistral Large 2 (Nov) Instruct (2411) | 59.23 | #465 |
| 11 | DeepSeek V4 Flash (Non-reasoning) | 59.21 | #137 (DeepSeek V4 Flash) |
| 12 | GLM-4.5 (Non-reasoning) | 58.64 | #275 (GLM-4.5) |
| 13 | DeepSeek V3.2 (Thinking) | 57.78 | #205 (DeepSeek V3.2) |
| 14 | DeepSeek V3.1 Terminus | 57.58 | #220 |
| 15 | DeepSeek V3 | 56.96 | #321 |
No result here: #8 Gemini 4 Argon, #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=ugi-leaderboard · How It Works · Data refreshed daily, snapshot 2026-10-20.