BehaviorBench - Personality from Demographics: leaderboard
Metric: Mean absolute error of the predicted score on one Big Five dimension from the demographics of the subject (Big Five personality dimension scores of held-out survey subjects); lower is better. Source: arxiv.org. Saturation forecast: Around August 2027. 18 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 6.4 |
| 2 | Llama 3.3 70B Instruct | 6.9 |
| 3 | GPT-4.1 | 6.9 |
| 4 | Claude Opus 4.6 | 7 |
| 5 | Gemini 3.1 Flash Lite (Preview) | 7.1 |
| 6 | GPT-5.4 Mini (High) | 7.2 |
| 7 | DeepSeek V3.2 | 7.3 |
| 8 | Qwen 3 4B | 7.3 |
| 9 | GPT-5.4 (High) | 7.4 |
| 10 | Claude Sonnet 4.6 | 7.6 |
| 11 | Claude Haiku 4.5 | 7.6 |
Interactive version: theaggregate.ai/benchmark?slug=behaviorbench-personality-from-demographics · How It Works · Data refreshed daily, snapshot 2026-09-29.