Best-action selection on Real-Store Pilot (n=20)
83.3Macro F1Claude Sonnet 4.6
Evaluation Results
| Method | Links | |
|---|---|---|
| Claude Sonnet 4.62026.06 | 83.3 | |
| Gemini 2.5 Flash2026.06 | 65.5 | |
| PIWM (full)2026.06 | 57.9 | |
| GPT-4o2026.06 | 53.5 | |
| State-Outcome Model2026.06 | 21.7 | |
| Qwen2.5-VL-7BProtocol=zero-shot2026.06 | 11.1 |