BehaviorBench (Behavioral Traces) - Trade Amount Error (ProfileGen): leaderboard

Metric: Median absolute error of the predicted trade amount (paper units, unbounded); history interface ProfileGen (a released structured profile summarizing the wallet's pre-target behavior); 2,000 pseudonymous prediction-market wallets reconstructed from public Polymarket metadata and Polygon on-chain logs, about 59,560 chronologically held-out test transactions; unparsable outputs count as failures; context-only baseline 31.79; lower is better. Source: arxiv.org. Saturation forecast: Around June 2027. 11 models tracked.

Top models

#ModelScore
1Llama 4 Scout Instruct10.3
2Gemma 3 12B (IT)10.99
3GPT-5.4 Mini12.96
4Claude 3.7 Sonnet14.12
5GPT-5.414.61
6Gemma 3 27B (IT)15.18
7Claude Opus 415.35
8GPT-OSS-20B15.5
9NVIDIA-Nemotron-3-Nano-30B-A3B-BF1617.07
10GPT-5.4 Nano20.83
11Ministral-3-14B-Instruct-251235.48

Interactive version: theaggregate.ai/benchmark?slug=behaviorbench-behavioral-traces-trade-amount-error-profilegen · How It Works · Data refreshed daily, snapshot 2026-09-29.