BehaviorBench (Behavioral Traces) - Trade Direction (RetrievalGen): leaderboard

Metric: Direction accuracy (%): predicted BUY or SELL of the next transaction; history interface RetrievalGen (behaviorally similar examples retrieved from a disjoint support pool of wallets); 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; chance 50, context-only baseline 52.85; higher is better. Source: arxiv.org. Saturation forecast: Around July 2027. 11 models tracked.

Top models

#ModelScore
1GPT-5.468.86
2Claude 3.7 Sonnet66.67
3GPT-5.4 Mini65.07
4Claude Opus 463.92
5Gemma 3 27B (IT)60.55
6Llama 4 Scout Instruct60.46
7NVIDIA-Nemotron-3-Nano-30B-A3B-BF1660.2
8GPT-OSS-20B59.99
9Gemma 3 12B (IT)59.82
10Ministral-3-14B-Instruct-251250.67
11GPT-5.4 Nano50.38

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