Lumen-bench - English: leaderboard
Lumen-bench harmful tool-action rate on the English versions of the 108 scenarios (both system prompts, 1,296 trials per model). It isolates conduct in one language where the headline pools all five.
Metric: Harmful tool-action rate (%, lower is better). Source: arxiv.org. 30 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 0 | #94 |
| 2 | GPT-5.4 | 0 | #123 |
| 3 | Claude Haiku 4.5 | 0 | #568 |
| 4 | GPT-5 Mini (2025-08-07) | 0 | #343 |
| 5 | MiniMax-M2.7 | 0.08 | #518 |
| 6 | Gemma 4 31B (IT) | 0.15 | #342 |
| 7 | Qwen 3.5 35B A3B | 0.15 | #482 |
| 8 | Qwen 3.5 397B A17B | 0.23 | #246 |
| 9 | Qwen 3.5 27B | 0.23 | #359 |
| 10 | Gemma 4 26B A4B (IT) | 0.23 | #549 |
| 11 | DeepSeek V3.2 | 0.39 | #406 |
| 12 | GPT-OSS-120B | 0.46 | #700 |
| 13 | GPT-4.1 | 0.62 | #495 |
| 14 | GPT-OSS-20B | 0.85 | #1147 |
| 15 | Qwen 3.5 9B | 2.39 | #803 |
Interactive version: theaggregate.ai/benchmark?slug=lumen-bench-english · How It Works · Data refreshed daily, snapshot 2026-11-13.