ANIMA
Models explain decisions about animal welfare. A judge scores their reasoning across 13 ethical dimensions.
The authors require access approval for the questions.
- Made by
- Compassion Aligned Machine Learning and Sentient Futures
- Released
- March 2026
- Kind
- Deception and alignment
- Level
- Everyday
- Scored
- Graded by a model
- Answers
- Open-ended
- Tasks
- All public
- Measures
- Capability
- Subject
- Philosophy, ethics and religion
- Language
- English
- Interaction
- Single turn
- Refresh
- One fixed set
- Run by
- Anyone
- Strongest model tested
- GPT-5.5 #21 on the capability index
- Last model added
- June 2026
- Scores refreshed
- November 20th, 2026
DeepSeek-V4-Flash leads with 74.4 among models released by June 2026
Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best.
The newest model with a score here came out in June 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 74.4
DeepSeek-V4-Flash
- 74.0 GPT-5.5
- 73.5 Claude Opus 4.7
- 72.5 Claude Opus 4.6
- 72.4
DeepSeek-V3.2
- 71.8 GPT-5 mini
- 71.0 GLM-5.2
- 70.8 GPT-5 nano
- 70.0
Gemini 3.5 Flash
- 69.1 GPT-5.2
- 61.6 Grok 4 Fast
- 58.6 Claude Haiku 4.5
- 58.4
Gemini 2.5 Flash-Lite
- 57.3 Claude 3 Opus
- 56.5 Claude Sonnet 4.6
- 52.7 Claude Sonnet 4.5
- 1
DeepSeek-V4-Flash 74.4
- 2 GPT-5.5 74.0
- 3 Claude Opus 4.7 73.5
- 4 Claude Opus 4.6 72.5
- 5
DeepSeek-V3.2 72.4
- 6 GPT-5 mini 71.8
- 7 GLM-5.2 71.0
- 8 GPT-5 nano 70.8
- 9
Gemini 3.5 Flash 70.0
- 10 GPT-5.2 69.1
- 11 Grok 4 Fast non-reasoning 61.6
- 12 Claude Haiku 4.5 58.6
- 13
Gemini 2.5 Flash-Lite Sep 2025 58.4
- 14 Claude 3 Opus 57.3
- 15 Claude Sonnet 4.6 56.5
- 16 Claude Sonnet 4.5 52.7
Source: Official ANIMA leaderboard
| # | Model | Score ± | Setting | Released | Run on | Source |
|---|---|---|---|---|---|---|
| 1 | | 74.4 ±4.7 | Apr 2026 | Jun 2026 | Official leaderboard | |
| 2 | GPT-5.5 | 74.0 ±4.8 | Apr 2026 | Apr 2026 | Official leaderboard | |
| 3 | Claude Opus 4.7 | 73.5 ±4.6 | Apr 2026 | Apr 2026 | Official leaderboard | |
| 4 | Claude Opus 4.6 | 72.5 ±5.1 | Feb 2026 | Feb 2026 | Official leaderboard | |
| 5 | | 72.4 ±5.3 | Dec 2025 | Feb 2026 | Official leaderboard | |
| 6 | GPT-5 mini | 71.8 ±4 | Aug 2025 | Jan 2026 | Official leaderboard | |
| 7 | GLM-5.2 | 71.0 ±4.7 | Jun 2026 | Jun 2026 | Official leaderboard | |
| 8 | GPT-5 nano | 70.8 ±4.2 | Aug 2025 | May 2026 | Official leaderboard | |
| 9 | | 70.0 ±4.9 | May 2026 | Jun 2026 | Official leaderboard | |
| 10 | GPT-5.2 | 69.1 ±5.7 | Dec 2025 | Feb 2026 | Official leaderboard | |
| 11 | Grok 4 Fast | 61.6 ±5.3 | non-reasoning | Sep 2025 | Jan 2026 | Official leaderboard |
| 12 | Claude Haiku 4.5 | 58.6 ±5.5 | Oct 2025 | Jan 2026 | Official leaderboard | |
| 13 | | 58.4 ±5.5 | Sep 2025 | Jan 2026 | Official leaderboard | |
| 14 | Claude 3 Opus | 57.3 ±5.5 | Feb 2024 | Mar 2026 | Official leaderboard | |
| 15 | Claude Sonnet 4.6 | 56.5 ±5.5 | Feb 2026 | Feb 2026 | Official leaderboard | |
| 16 | Claude Sonnet 4.5 | 52.7 ±5.5 | Sep 2025 | Feb 2026 | Official leaderboard |
Rankings by part
We read 13 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Actionability, Cautious Impact Consideration and Contextual Welfare Salience. DeepSeek-V4-Flash leads 3 of them, and 7 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Actionability
16 models we track have a score here.
- 1 GPT-5.2 ↑9 92.0
- 2 GPT-5.5 88.0
- 3 GPT-5 nano ↑5 86.0
- 4 GPT-5 mini ↑2 83.0
- 5
DeepSeek-V4-Flash ↓4 82.0
- 6
DeepSeek-V3.2 ↓1 77.0
- 7 GLM-5.2 77.0
- 8 Claude Opus 4.7 ↓5 73.0
- 9
Gemini 3.5 Flash 71.0
- 10 Claude Haiku 4.5 ↑2 70.0
Cautious Impact Consideration
16 models we track have a score here.
- 1
DeepSeek-V4-Flash 88.0
- 2
DeepSeek-V3.2 ↑3 85.0
- 3 Claude Opus 4.6 ↑1 85.0
- 4 Claude Opus 4.7 ↓1 85.0
- 5
Gemini 3.5 Flash ↑4 84.0
- 6 GPT-5.5 ↓4 82.0
- 7 GLM-5.2 80.0
- 8 GPT-5.2 ↑2 77.0
- 9 Grok 4 Fast ↑2 75.0
- 10 Claude 3 Opus ↑4 74.0
Contextual Welfare Salience
16 models we track have a score here.
- 1 Claude Opus 4.6 ↑3 84.0
- 2 Claude Opus 4.7 ↑1 84.0
- 3 GPT-5.5 ↓1 84.0
- 4 GPT-5 mini ↑2 83.0
- 5
DeepSeek-V3.2 83.0
- 6 GPT-5 nano ↑2 81.0
- 7
DeepSeek-V4-Flash ↓6 80.0
- 8 GLM-5.2 ↓1 79.0
- 9
Gemini 3.5 Flash 78.0
- 10
Gemini 2.5 Flash-Lite Sep 2025 ↑3 71.0
Control Questions
16 models we track have a score here.
- 1 Claude Sonnet 4.5 ↑15 100.0
- 2 Claude Haiku 4.5 ↑10 100.0
- 3 Claude Opus 4.6 ↑1 100.0
- 4 Claude Sonnet 4.6 ↑11 100.0
- 5 GPT-5 nano ↑3 99.0
- 6 Grok 4 Fast ↑5 99.0
- 7
Gemini 2.5 Flash-Lite Sep 2025 ↑6 99.0
- 8 GPT-5 mini ↓2 95.0
- 9 GPT-5.5 ↓7 94.0
- 10 Claude Opus 4.7 ↓7 93.0
Epistemic Humility
16 models we track have a score here.
- 1 Claude Opus 4.6 ↑3 95.0
- 2 Claude Opus 4.7 ↑1 93.0
- 3 GPT-5.5 ↓1 92.0
- 4 GPT-5 nano ↑4 84.0
- 5
DeepSeek-V3.2 80.0
- 6 GLM-5.2 ↑1 80.0
- 7
DeepSeek-V4-Flash ↓6 73.0
- 8
Gemini 3.5 Flash ↑1 73.0
- 9 GPT-5 mini ↓3 71.0
- 10 Claude 3 Opus ↑4 68.0
Evidence-Based Capacity Attribution
16 models we track have a score here.
- 1 Grok 4 Fast ↑10 39.0
- 2 Claude Opus 4.7 ↑1 39.0
- 3 GPT-5 mini ↑3 38.0
- 4 GPT-5.2 ↑6 38.0
- 5
DeepSeek-V3.2 35.0
- 6 GPT-5 nano ↑2 33.0
- 7 Claude Haiku 4.5 ↑5 32.0
- 8 Claude Sonnet 4.6 ↑7 31.0
- 9
DeepSeek-V4-Flash ↓8 31.0
- 10
Gemini 2.5 Flash-Lite Sep 2025 ↑3 27.0
Harm Minimization
16 models we track have a score here.
- 1 GPT-5 mini ↑5 90.0
- 2 GPT-5.5 87.0
- 3 GPT-5 nano ↑5 85.0
- 4
DeepSeek-V4-Flash ↓3 82.0
- 5
DeepSeek-V3.2 80.0
- 6 GPT-5.2 ↑4 80.0
- 7 Claude Opus 4.7 ↓4 80.0
- 8 Claude Opus 4.6 ↓4 79.0
- 9 GLM-5.2 ↓2 76.0
- 10
Gemini 3.5 Flash ↓1 74.0
Moral Consideration
16 models we track have a score here.
- 1
DeepSeek-V4-Flash 57.0
- 2
DeepSeek-V3.2 ↑3 55.0
- 3 GLM-5.2 ↑4 52.0
- 4 Claude Opus 4.7 ↓1 51.0
- 5
Gemini 3.5 Flash ↑4 51.0
- 6 Claude Opus 4.6 ↓2 50.0
- 7 GPT-5.5 ↓5 48.0
- 8 GPT-5.2 ↑2 47.0
- 9 GPT-5 nano ↓1 41.0
- 10
Gemini 2.5 Flash-Lite Sep 2025 ↑3 39.0
Novel Entity Precaution
16 models we track have a score here.
- 1
DeepSeek-V3.2 ↑4 60.0
- 2 Claude Opus 4.6 ↑2 53.0
- 3
DeepSeek-V4-Flash ↓2 43.0
- 4 GPT-5 mini ↑2 40.0
- 5 GPT-5.5 ↓3 39.0
- 6 Claude Opus 4.7 ↓3 38.0
- 7 GPT-5.2 ↑3 35.0
- 8 Grok 4 Fast ↑3 34.0
- 9
Gemini 2.5 Flash-Lite Sep 2025 ↑4 33.0
- 10 GPT-5 nano ↓2 30.0
Prejudice Avoidance
16 models we track have a score here.
- 1
DeepSeek-V3.2 ↑4 100.0
- 2 Claude Opus 4.6 ↑2 99.0
- 3 GLM-5.2 ↑4 99.0
- 4 Claude Opus 4.7 ↓1 98.0
- 5 GPT-5.5 ↓3 98.0
- 6
Gemini 3.5 Flash ↑3 98.0
- 7 Claude 3 Opus ↑7 97.0
- 8 GPT-5.2 ↑2 97.0
- 9
DeepSeek-V4-Flash ↓8 97.0
- 10 GPT-5 nano ↓2 95.0
Scope Sensitivity
16 models we track have a score here.
- 1 Claude Opus 4.7 ↑2 54.0
- 2
DeepSeek-V4-Flash ↓1 54.0
- 3 GPT-5 mini ↑3 53.0
- 4 Grok 4 Fast ↑7 53.0
- 5 GLM-5.2 ↑2 53.0
- 6
Gemini 3.5 Flash ↑3 50.0
- 7
DeepSeek-V3.2 ↓2 49.0
- 8 Claude Opus 4.6 ↓4 44.0
- 9 Claude Haiku 4.5 ↑3 40.0
- 10 GPT-5.5 ↓8 40.0
Sentience Acknowledgement
16 models we track have a score here.
- 1 Claude Opus 4.6 ↑3 65.0
- 2 Claude Opus 4.7 ↑1 55.0
- 3
DeepSeek-V4-Flash ↓2 50.0
- 4 GPT-5 mini ↑2 46.0
- 5 GPT-5.5 ↓3 45.0
- 6 Claude Sonnet 4.6 ↑9 44.0
- 7 GLM-5.2 40.0
- 8 GPT-5 nano 38.0
- 9
Gemini 3.5 Flash 38.0
- 10 Grok 4 Fast ↑1 37.0
Trade-Off Transparency
16 models we track have a score here.
- 1
DeepSeek-V4-Flash 81.0
- 2
Gemini 3.5 Flash ↑7 76.0
- 3 Claude Opus 4.6 ↑1 75.0
- 4 Claude Opus 4.7 ↓1 75.0
- 5 GLM-5.2 ↑2 73.0
- 6 GPT-5 mini 70.0
- 7 GPT-5.5 ↓5 69.0
- 8
Gemini 2.5 Flash-Lite Sep 2025 ↑5 68.0
- 9 Grok 4 Fast ↑2 65.0
- 10
DeepSeek-V3.2 ↓5 65.0
4 models have held number one on the official leaderboard since its release in March 2026
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official ANIMA leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official ANIMA leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Safety and alignment
- FORTRESS In use Paired adversarial and benign national-security misuse prompts. Rubric-graded risk score on adversarial and over-refusal score on benign prompts. Claude Opus 5.5 5.78%
- StrongREJECT In use Forbidden-prompt jailbreak benchmark with a rubric grader that scores how useful a non-refusal actually is.
Inkling 98.6%
- Gray Swan IPI Arena benchmark In use Indirect prompt-injection attacks from public red-teaming competitions replayed on tool, coding and browser agents. Attack success rate at k attempts.
Gemini 4 Carbon 0%
- Covert misbehavior (Thinking Machines) In use Agentic tasks that check whether a model misbehaves and conceals it. Scored by the rate of covert misbehavior.
Inkling 2 Mini 8.30%
- Destructive actions (Thinking Machines) In use Agentic tasks that check whether a model takes destructive actions it was not authorized to take. Scored by the rate of destructive actions.
Inkling 2 Small 4.70%
- Misbehavior (Thinking Machines) In use Agentic tasks that check whether a model fabricates results, ignores instructions or hides what it did when a task is hard. Scored by the rate of misbehavior.
Inkling 2 Mini 9.80%
Data audit
All 9 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (16 entries).
- Pass Models scored 16 models we track.
- Pass Newest score The newest is from June 2026, for GLM-5.2.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.5 (#21).
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on compassionbench.com: 16 rows for models we track at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.