Visual Puzzle Reasoning on PuzzleVQA
46.8AccuracyPEPOD
Evaluation Results
| Method | Links | |
|---|---|---|
| PEPODBackbone=Qwen2.5-VL-3B-Instruct2026.03 | 46.8 | |
| PEPOGBackbone=InternVL3-2B-Instruct2026.03 | 45.2 | |
| PEPODBackbone=InternVL3-2B-Instruct2026.03 | 45.2 | |
| PEPOGBackbone=Qwen2.5-VL-3B-Instruct2026.03 | 45 | |
| DAPOBackbone=InternVL3-2B-Instruct2026.03 | 44.8 | |
| DAPOBackbone=Qwen2.5-VL-3B-Instruct2026.03 | 44.6 | |
| GRPOBackbone=Qwen2.5-VL-3B-Instruct2026.03 | 43.2 | |
| GRPOBackbone=InternVL3-2B-Instruct2026.03 | 43.2 | |
| High-Entropy RLBackbone=Qwen2.5-VL-3B-Instruct2026.03 | 35 | |
| High-Entropy RLBackbone=InternVL3-2B-Instruct2026.03 | 32.6 | |
| Base (zero-shot)Backbone=Qwen2.5-VL-3B-Instruct2026.03 | 31.8 | |
| Base (zero-shot)Backbone=InternVL3-2B-Instruct2026.03 | 31.8 |