ANIMA (AnimalHarmBench 2.0): leaderboard
ANIMA (Animal Norms In Moral Assessment; CaML and AI for Animals, formerly AnimalHarmBench 2.0): open-ended questions, many in languages other than English, about situations with consequences for animals. Several LLM judges grade each answer on 13 moral-reasoning dimensions such as moral consideration, harm minimization, sentience acknowledgement, scope sensitivity and epistemic humility, with control questions that penalise needless moralising. Score: overall mean (0-100) across dimensions and epochs. Higher is better. Source: https://compassionbench.com/
Metric: Overall score (0-100; multi-judge LLM grading of reasoning about harm to animals, mean over 13 moral-reasoning dimensions and epochs). Source: compassionbench.com. Saturation forecast: Around December 2027. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V4 Flash | 74.4 | #136 |
| 2 | GPT-5.5 | 74 | #31 |
| 3 | Claude Opus 4.7 | 73.5 | #47 |
| 4 | Claude Opus 4.6 | 72.5 | #62 |
| 5 | DeepSeek V3.2 | 72.4 | #201 |
| 6 | GPT-5 Mini | 71.8 | #183 |
| 7 | GLM-5.2 | 71 | #72 |
| 8 | GPT-5 Nano | 70.8 | #423 |
| 9 | Gemini 3.5 Flash | 70 | #59 |
| 10 | GPT-5.2 | 69.1 | #109 |
| 11 | Grok 4 Fast (Non-reasoning) | 61.6 | #252 (Grok 4 Fast) |
| 12 | Claude Haiku 4.5 | 58.6 | #278 |
| 13 | Gemini 2.5 Flash Lite | 58.4 | #420 |
| 14 | Claude 3 Opus | 57.3 | #570 |
| 15 | Claude Sonnet 4.6 | 56.5 | #90 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=anima-animalharmbench-2-0 · How It Works · Data refreshed daily, snapshot 2026-10-19.