Multimodal mathematical reasoning on MathVis.m
47.37AccuracyCEPO
Evaluation results
| Method | Links | |
|---|---|---|
| CEPOBackbone=Qwen3-VL-4B-Instruct2026.05 | 47.37 | |
| BaseBackbone=Qwen3-VL-4B-Instruct2026.05 | 44.41 | |
| OPSDBackbone=Qwen3-VL-4B-Instruct2026.05 | 44.41 | |
| RLSDBackbone=Qwen3-VL-4B-Instruct2026.05 | 44.08 | |
| SDPOBackbone=Qwen3-VL-4B-Instruct2026.05 | 43.42 | |
| GRPOBackbone=Qwen3-VL-4B-Instruct2026.05 | 42.76 | |
| CEPOBackbone=Qwen3-VL-2B-Instruct2026.05 | 25.99 | |
| RLSDBackbone=Qwen3-VL-2B-Instruct2026.05 | 23.39 | |
| GRPOBackbone=Qwen3-VL-2B-Instruct2026.05 | 21.05 | |
| BaseBackbone=Qwen3-VL-2B-Instruct2026.05 | 19.41 | |
| SDPOBackbone=Qwen3-VL-2B-Instruct2026.05 | 15.46 | |
| OPSDBackbone=Qwen3-VL-2B-Instruct2026.05 | 14.14 |