CrueltyBench: leaderboard
CrueltyBench (CaML): single-turn, realistic in-persona requests with implicit animal-cruelty implications that never mention the animal's welfare (18 harm scenarios plus controls, 5 epochs). A fixed two-judge panel (Claude Opus 5, Gemini 3.7 Flash) grades each response green (clearly reduces or mitigates the cruelty), yellow or red; control items penalise moralising at ordinary requests. Score: mean welfare score (0-100; green 1, yellow 0.5, red 0). Higher is better. Source: https://compassionbench.com/crueltybench
Metric: Welfare score (0-100; two-judge panel, green 1 / yellow 0.5 / red 0, mean over the implicit animal-cruelty requests and epochs). Source: compassionbench.com. Saturation forecast: Around September 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra | 81.7 | #6 |
| 2 | GPT-5.6 Sol | 72.8 | #20 |
| 3 | GPT-5.5 | 63.6 | #31 |
| 4 | Claude Opus 5 | 56.7 | #12 |
| 5 | Muse Spark 1.3 | 51.7 | #25 |
| 6 | Claude Opus 5.5 | 49.7 | #3 |
| 7 | Qwen 3.8 Max | 49.4 | #32 |
| 8 | Grok 4.6 | 49.2 | #26 |
| 9 | Claude Fable 5.1 | 39.7 | #9 |
| 10 | Claude Sonnet 5.5 | 37.5 | #11 |
| 11 | Gemini 3.8 Flash | 35 | #22 |
| 12 | Kimi K3 | 32.5 | #30 |
| 13 | Claude Sonnet 5 | 26.4 | #56 |
| 14 | DeepSeek V4 Pro | 23.9 | #100 |
No result here: #7 GPT-6.1 Sol, #8 Gemini 4 Argon, #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=crueltybench · How It Works · Data refreshed daily, snapshot 2026-10-19.