EvalMORAAL: leaderboard
The model rates, with a chain-of-thought prompt, how acceptable 19 contested behaviours (abortion, bribery, euthanasia and others) are in each of 55 countries; the score is the Pearson correlation with the World Values Survey 2017-2022 country averages. It tests how well a model represents how moral attitudes vary between countries, not whether its own verdicts are right.
Metric: Pearson r between direct chain-of-thought ratings and World Values Survey country averages. Source: arxiv.org. Saturation forecast: Estimated already saturated. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3 Opus (20240229) | 0.9 | #1103 |
| 2 | GPT-4o (2024-05-13) | 0.89 | #735 |
| 3 | Gemini 1.0 Pro | 0.89 | #4043 |
| 4 | GPT-4 (0613) | 0.85 | #1214 |
| 5 | GPT-4o Mini (2024-07-18) | 0.84 | #1510 |
| 6 | Phi-3 Mini 4K Instruct | 0.83 | #6810 |
| 7 | Mistral Large 2 (Jul) | 0.81 | #1155 |
| 8 | Mistral 7B Instruct (v0.3) | 0.77 | #6653 |
| 9 | Gemini 2.0 Flash (Preview) | 0.77 | #736 |
| 10 | O1 Preview (2024-09-12) | 0.77 | #391 |
| 11 | Llama 3.3 70B Instruct | 0.75 | #1357 |
| 12 | Claude 3 Sonnet (20240229) | 0.73 | #1911 |
| 13 | Llama 3.2 3B Instruct | 0.73 | #7147 |
| 14 | Command-R+ (08-2024) | 0.72 | #2551 |
| 15 | GPT-3.5 Turbo (0125) | 0.7 | #3500 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=evalmoraal · How It Works · Data refreshed daily, snapshot 2026-11-13.