DystopiaBench: leaderboard
DystopiaBench — measures compliance with dystopian/authoritarian misuse requests; Dystopian Compliance Score 0-100 (lower compliance is safer behaviour).
Metric: Dystopian Compliance Score (0-100). Source: dystopiabench.com. Saturation forecast: Not forecast. 42 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Haiku 4.5 | 23.8 | #280 |
| 2 | Claude Opus 4.7 | 25.2 | #49 |
| 3 | Claude Sonnet 4.6 | 25.4 | #90 |
| 4 | Claude Opus 4.6 | 28.1 | #63 |
| 5 | GLM-5.1 | 29.6 | #97 |
| 6 | GLM-5 | 31.7 | #141 |
| 7 | MiMo-V2.5 | 34.5 | #196 |
| 8 | MiMo-V2.5-Pro | 36 | #158 |
| 9 | GPT-5.4 | 43.5 | #79 |
| 10 | GPT-5.4 Mini | 44.5 | #212 |
| 11 | MiniMax-M2.7 | 47.5 | #258 |
| 12 | GPT-5.5 | 51.2 | #31 |
| 13 | MiniMax-M2.5 | 51.5 | #303 |
| 14 | GPT-5.3 Codex | 54.7 | #69 |
| 15 | Gemini 3.1 Pro (Preview) | 57.7 | #58 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=dystopiabench · How It Works · Data refreshed daily, snapshot 2026-10-20.