Visual Mathematical Reasoning on FlowVerse DI+RP+EI+OQ (All)
34.4CoT ErrorQwen-VL-Plus
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Qwen-VL-Plus2025.03 | 34.4 | 27.1 | |
| InfiMM-Math-7BModel size=7B2025.03 | 36.5 | 28.8 | |
| MathLLaVA-13BModel size=13B2025.03 | 38 | 29.9 | |
| Gemini-Pro2025.03 | 39 | 23.5 | |
| Qwen-VL-Max2025.03 | 43 | 36.3 | |
| MultiMath-7BModel size=7B2025.03 | 44 | 34.2 | |
| InternVL2.5-8BModel size=8B2025.03 | 44.7 | 41 | |
| SVE-Math-Qwen2.5-7BBackbone=Qwen2.5, Model size=7B2025.03 | 46 | 39.4 | |
| VLM-R1-7B†Model size=7B2025.03 | 50.7 | 41.2 | |
| GPT-4o-mini2025.03 | 51.3 | 44.5 | |
| Qwen2-VL-72BModel size=72B2025.03 | 52.3 | 48.6 | |
| Qwen2.5-VL-7BModel size=7B2025.03 | 53.8 | 42.2 | |
| InternVL2.5-78BModel size=78B2025.03 | 54.7 | 50.1 | |
| GPT-4o2025.03 | 55.1 | 47.8 | |
| GPT-4V2025.03 | 56.2 | 53.4 | |
| Claude-3.5-Sonnet2025.03 | 56.9 | 49.6 | |
| MathFlowBackbone=Qwen2.5-VL-7B2025.03 | 57 | 46 | |
| Gemini-2.5-pro2025.03 | 62 | 55.3 | |
| GPT-52025.03 | 65.8 | 60.1 | |
| MathFlowBackbone=GPT-52025.03 | 66.5 | 61.8 |