BehaviorBench (Behavioral Traces) - Belief Confidence Error (ProfileGen): leaderboard
BehaviorBench (Yang et al., 2026) tests whether a model can predict what a real pseudonymous user will do, using 2,000 wallets reconstructed from public prediction-market and on-chain records with chronologically held-out targets. The model sees a structured profile summarizing the past behavior of the wallet (ProfileGen). The board is the mean absolute error, on a 0 to 1 scale, of the predicted share of the final position of the user held on that side; lower is better.
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 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, 22,204 chronologically held-out wallet-market test instances; context-only baseline 0.3417; lower is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.7 Sonnet | 0.08 | #458 |
| 2 | Claude Opus 4 | 0.08 | #322 |
| 3 | Gemma 3 12B (IT) | 0.08 | #1837 |
| 4 | Gemma 3 27B (IT) | 0.09 | #1289 |
| 5 | GPT-OSS-20B | 0.1 | #1147 |
| 6 | GPT-5.4 Mini | 0.13 | #389 |
| 7 | Llama 4 Scout Instruct | 0.13 | #1543 |
| 8 | GPT-5.4 Nano | 0.15 | #635 |
| 9 | GPT-5.4 | 0.18 | #123 |
| 10 | Ministral-3-14B-Instruct-2512 | 0.19 | #1593 |
| 11 | NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 | 0.2 | #1767 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=behaviorbench-behavioral-traces-belief-confidence-error-profilegen · How It Works · Data refreshed daily, snapshot 2026-11-13.