Next-step Generation on Human Evaluation Subset
65Win RateCoP (Qwen2.5-VL-7B)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CoP (Qwen2.5-VL-7B)Backbone=Qwen2.5-VL-7B2026.05 | 65 | 2 | 33 | |
| CoP (GPT-4o)Backbone=GPT-4o2026.05 | 48 | 20 | 31 |
| Method | Links | |||
|---|---|---|---|---|
| CoP (Qwen2.5-VL-7B)Backbone=Qwen2.5-VL-7B2026.05 | 65 | 2 | 33 | |
| CoP (GPT-4o)Backbone=GPT-4o2026.05 | 48 | 20 | 31 |