Diagram Understanding on AI2D (Pass@1 accuracy)
86.5Pass@1 AccuracyQwen2.5-VL-7B-Instruct + cold start
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2.5-VL-7B-Instruct + cold startModel Source Category=Ours, Training Strategy=cold start2026.04 | 86.5 | |
| InternVL2.5-26BModel Source Category=Open-Source Models2026.04 | 86.4 | |
| LLaVA-OneVision-72BModel Source Category=Open-Source Models2026.04 | 85.6 | |
| GPT-4o-20240513Model Source Category=Closed-Source Models2026.04 | 84.6 | |
| InternVL3-9BModel Source Category=Open-Source Models2026.04 | 84.6 | |
| InternVL2.5-8BModel Source Category=Open-Source Models2026.04 | 84.5 | |
| Qwen2.5-VL-7B-Instruct + RFTModel Source Category=Ours, Training Strategy=RFT2026.04 | 84.1 | |
| Qwen2.5-VL-7B-InstructModel Source Category=Base Model2026.04 | 83.9 | |
| Qwen2.5-VL-7B-InstructModel Source Category=Ours2026.04 | 83.9 | |
| Qwen2.5-VL-7B-InstructModel Source Category=Base Model, Chain-of-Thought (CoT)=true2026.04 | 83.1 | |
| Qwen2-VL-7BModel Source Category=Open-Source Models2026.04 | 83 | |
| MiniCPM-V2.6Model Source Category=Open-Source Models2026.04 | 82.1 | |
| Claude-3.5-SonnetModel Source Category=Closed-Source Models2026.04 | 81.2 | |
| Cambrian-34BModel Source Category=Open-Source Models2026.04 | 79.5 | |
| Gemini-1.5-ProModel Source Category=Closed-Source Models2026.04 | 79.1 | |
| GPT-4VModel Source Category=Closed-Source Models2026.04 | 78.2 |