BehaviorBench (Behavioral Traces) - Belief Confidence Error (RetrievalGen): leaderboard

Metric: Confidence mean absolute error on a 0 to 1 scale: gap between the predicted and revealed share of the final position held on the dominant side; 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, 22,204 chronologically held-out wallet-market test instances; context-only baseline 0.3417; lower is better. Source: arxiv.org. Saturation forecast: Around December 2026. 11 models tracked.

Top models

#ModelScore
1Claude Opus 40.11
2Gemma 3 12B (IT)0.13
3NVIDIA-Nemotron-3-Nano-30B-A3B-BF160.17
4Gemma 3 27B (IT)0.19
5Claude 3.7 Sonnet0.2
6Llama 4 Scout Instruct0.2
7GPT-OSS-20B0.25
8Ministral-3-14B-Instruct-25120.25
9GPT-5.4 Mini0.27
10GPT-5.4 Nano0.3
11GPT-5.40.31

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