Price Scenario Prediction on S&P 500 2017 (held-out)
27.3Top-1 Scenario AccuracyClaude Sonnet 3.7
Evaluation results
| Method | Links | |
|---|---|---|
| Claude Sonnet 3.7Evaluation Protocol=zero-shot2026.04 | 27.3 | |
| Hindsight Preference Optimization (HPO)Backbone=Qwen3-VL-4B, Evaluation Protocol=Hindsight DPO2026.04 | 27.1 | |
| Qwen3-VL-4BEvaluation Protocol=SFT (Stage 1)2026.04 | 25.2 | |
| Qwen3-VL-235BEvaluation Protocol=zero-shot2026.04 | 23.7 | |
| Qwen3-VL-4BEvaluation Protocol=zero-shot2026.04 | 22.1 |