Multimodal Mathematical Reasoning on MathVista
85.1GPS ScoreClaude-4-Sonnet
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Claude-4-SonnetModel Category=Proprietary2026.02 | 85.1 | 81.9 | |
| Gemini-2.5-ProModel Category=Proprietary2026.02 | 79.33 | 79 | |
| Guided-GRPOBackbone=Qwen3-8B, Training=G-GRPO+SFT2026.02 | 77.88 | 76.51 | |
| Vision-R1-32BModel Category=Open-Source2026.02 | 75 | 74.73 | |
| Qwen2.5-VL-32BModel Category=Open-Source2026.02 | 70.67 | 61.57 | |
| GPT-4oModel Category=Proprietary2026.02 | 69.23 | 67.97 | |
| Qwen-VL-MaxModel Category=Proprietary2026.02 | 68.75 | 58.36 | |
| Base (Qwen3-8B)Backbone=Qwen3-8B2026.02 | 67.31 | 64.06 | |
| VL-Rethinker-7BModel Category=Open-Source2026.02 | 66.93 | 65.12 | |
| InternVL2.5-26BModel Category=Open-Source2026.02 | 65.38 | 53.38 | |
| Llama-3.2-11BModel Category=Open-Source2026.02 | 60.58 | 55.16 | |
| LLaVA-v1.5-13BModel Category=Open-Source2026.02 | 57.69 | 51.6 | |
| MMR1-32BModel Category=Open-Source2026.02 | 43.75 | 44.48 | |
| Math-LLaVA-13BModel Category=Open-Source2026.02 | 42.79 | 39.86 | |
| Phi-3-vision-128kModel Category=Open-Source2026.02 | 35.1 | 33.45 |