Chart Understanding on CharXiv
60.2Reasoning ScoreClaude Sonnet 3.5
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Claude Sonnet 3.5Param.=–2026.04 | 60.2 | 84.3 | |
| CHARTOOL-7BParam.=7B2026.04 | 50.5 | 80.3 | |
| Qwen2.5-VL-72BParam.=72B2026.04 | 49.7 | 87.4 | |
| DeepEyesV2Tool=General, Param Size=7B2025.11 | 48.9 | 78.6 | |
| Deepeyes-V2Param.=7B2026.04 | 48.9 | 78.6 | |
| Qwen2.5-VLTool=✗, Param Size=32B2025.11 | 48 | 83.2 | |
| GPT-4oParam.=–2026.04 | 47.1 | 84.45 | |
| START-RLModel size=7B2025.12 | 46.7 | 77.6 | |
| Qwen3-VL-8BParam.=8B2026.04 | 46.4 | 83 | |
| Chart-R1Model size=7B2025.12 | 45.2 | 62 | |
| GPT-4oTool use=No2025.11 | 45 | 85.4 | |
| START-SFTModel size=7B2025.12 | 44 | 66.9 | |
| Qwen2.5-VL-7BParam.=7B2026.04 | 42.5 | 73.6 | |
| ThymeParam.=7B2026.04 | 42.5 | 65.33 | |
| ECD Qwen2.5-VL-7BParam.=7B2026.04 | 41.8 | 75.08 | |
| Qwen2.5-VLModel size=7B2025.12 | 41.4 | 66.6 | |
| DeepeyesParam.=7B2026.04 | 41.4 | 64.95 | |
| CHARTOOL-3BParam.=3B2026.04 | 41.4 | 74.75 | |
| ECDModel size=7B2025.12 | 40.2 | 74.2 | |
| Qwen2.5-VLTool=✗, Param Size=7B2025.11 | 40.2 | 72.7 | |
| START-RLModel size=3B2025.12 | 40 | 72.2 | |
| InternVL3-8BParam.=8B2026.04 | 39.5 | 75.33 | |
| CodeV-7B-RLTool use=Yes, Training=RL2025.11 | 39.3 | 72.1 | |
| Pixel-Reasoner-7BTool use=Yes2025.11 | 38.8 | 70.7 | |
| ThinkLite-7BTool use=No2025.11 | 38.4 | 75.9 | |
| Thyme-RL-7BTool use=Yes2025.11 | 38.2 | 67.4 | |
| InternVL3-8BTool use=No2025.11 | 37.8 | 72.2 | |
| InternVL3Tool=✗, Param Size=8B2025.11 | 37.6 | 73.6 | |
| DeepEyes-7BTool use=Yes2025.11 | 36.7 | 70.5 | |
| Qwen2.5-VL-7BTool use=No2025.11 | 36.3 | 71.8 | |
| START-SFTModel size=3B2025.12 | 35.9 | 63.1 | |
| Penguin-VLParameters=2B2026.03 | 35.8 | 66.4 | |
| CodeV-7B-SFTTool use=Yes, Training=SFT2025.11 | 32.3 | 70.6 | |
| Qwen2.5-VLModel size=3B2025.12 | 32.1 | 59.7 | |
| InternVL3.5Parameters=2B2026.03 | 31.6 | 65 | |
| Qwen2.5-VL-3BParam.=3B2026.04 | 31.3 | 58.6 | |
| ChartMOEParam.=8B2026.04 | 28.3 | 47.42 | |
| Gemma3n E2B-itParameters=E2B-it, Inference template=Penguin's template2026.03 | 27 | 60.1 | |
| Qwen3-VLParameters=2B2026.03 | 26.8 | 62.3 | |
| ChartGemmaParam.=3B2026.04 | 21.6 | 15.88 | |
| SmolVLM2Parameters=2.2B2026.03 | 15.5 | 36.9 | |
| ChartGemmaModel size=3B2025.12 | 12.5 | — | |
| TinyChartParam.=3B2026.04 | 12.2 | 16.12 | |
| TinyChartModel size=3B2025.12 | 8.3 | — |