BehaviorBench - Multi-Round Game Prediction: leaderboard

Metric: Given the game rules and the earlier rounds of one subject, predict the next move, on MobLab economic-game records of human players (dictator, ultimatum, trust, public goods, bomb risk, beauty contest and push/pull games): Mean absolute error of the predicted move in the game choice units; lower is better. Source: arxiv.org. Saturation forecast: Around August 2027. 18 models tracked.

Top models

#ModelScore
1GPT-4.118.1
2Claude Sonnet 4.618.7
3Claude Opus 4.619.3
4Gemini 3.1 Pro (Preview)19.4
5DeepSeek V3.220.9
6Claude Haiku 4.521
7GPT-5.4 (High)21.1
8Qwen 3 4B21.7
9GPT-5.4 Mini (High)23.7
10Gemini 3.1 Flash Lite (Preview)25.5
11Llama 3.3 70B Instruct27

Interactive version: theaggregate.ai/benchmark?slug=behaviorbench-multi-round-game-prediction · How It Works · Data refreshed daily, snapshot 2026-09-29.