BehaviorBench - Cross-Game Prediction (Distribution): leaderboard
BehaviorBench (University of Michigan, 2026) evaluates foundation models on behavioral science tasks at the individual and population levels, using economic-game records, Big Five personality surveys and research literature. This board is the Wasserstein distance (0-100) between predicted and observed first-move distributions when predicting across games; lower is better.
Metric: Given the first-round moves of each subject in other games, predict the first-round moves in a target game, on MobLab economic-game records of human players (dictator, ultimatum, trust, public goods, bomb risk, beauty contest and push/pull games): Wasserstein distance (0-100) between the model-predicted and the observed human choice distributions, with choices normalized to 0-100 and averaged over games; lower is better. Source: arxiv.org. Saturation forecast: Around February 2027. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 10.3 | #94 |
| 2 | DeepSeek V3.2 | 12.3 | #406 |
| 3 | GPT-5.4 (High) | 14 | #123 (GPT-5.4) |
| 4 | Llama 3.3 70B Instruct | 14.9 | #1357 |
| 5 | GPT-4.1 | 16.3 | #495 |
| 6 | Claude Haiku 4.5 | 17.3 | #568 |
| 7 | Gemini 3.1 Flash Lite (Preview) | 17.3 | #418 |
| 8 | GPT-5.4 Mini (High) | 18 | #389 (GPT-5.4 Mini) |
| 9 | Claude Sonnet 4.6 | 19.3 | #161 |
| 10 | Claude Opus 4.6 | 20.1 | #102 |
| 11 | Qwen 3 4B | 20.1 | #2281 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=behaviorbench-cross-game-prediction-distribution · How It Works · Data refreshed daily, snapshot 2026-11-13.