BehaviorBench - Multi-Round 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 next-move distributions in multi-round games, with choices normalized to 0-100; lower is better.
Metric: Given the game rules and the earlier rounds of each subject, predict the next moves, 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: Estimated already saturated. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 | 6.9 | #495 |
| 2 | DeepSeek V3.2 | 10.1 | #406 |
| 3 | Claude Sonnet 4.6 | 10.6 | #161 |
| 4 | Claude Opus 4.6 | 11.7 | #102 |
| 5 | Gemini 3.1 Pro (Preview) | 12.5 | #94 |
| 6 | Claude Haiku 4.5 | 12.5 | #568 |
| 7 | Qwen 3 4B | 13.1 | #2281 |
| 8 | GPT-5.4 (High) | 15.2 | #123 (GPT-5.4) |
| 9 | Llama 3.3 70B Instruct | 15.3 | #1357 |
| 10 | GPT-5.4 Mini (High) | 17.6 | #389 (GPT-5.4 Mini) |
| 11 | Gemini 3.1 Flash Lite (Preview) | 18.6 | #418 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=behaviorbench-multi-round-game-prediction-distribution · How It Works · Data refreshed daily, snapshot 2026-11-13.