Chart-based Question Answering on ChartQA Pro
76.9AccuracyClaude Sonnet 3.7
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Claude Sonnet 3.7Model Category=SOTA MLLM (Proprietary)2026.04 | 76.9 | — | |
| Claude Sonnet 4.5Model Category=SOTA MLLM (Proprietary)2026.04 | 75 | — | |
| Qwen3-VL-4B-Instruct-DAPOModel Category=Fine-tuned VLM, Policy Optimization=DAPO2026.04 | 63.4 | 9.48 | |
| Qwen3-VL-4B-Instruct-GRPOModel Category=Fine-tuned VLM, Policy Optimization=GRPO2026.04 | 62.7 | 9.84 | |
| Qwen3-VL-4B-Instruct-GSPOModel Category=Fine-tuned VLM, Policy Optimization=GSPO2026.04 | 62.2 | 9.69 | |
| Qwen3-VL-8B-InstructModel Category=Open-source VLM2026.04 | 58 | 31.59 | |
| Llava-v1.6Model Category=Open-source VLM2026.04 | 57 | 6.19 | |
| SketchVL-7BModel Scale=7B, Training Strategy=FinePO2026.01 | 52.62 | — | |
| Qwen2.5VL-7BModel Scale=7B2026.01 | 52.4 | — | |
| Qwen2.5-VL-7B-InstructModel Category=Open-source VLM2026.04 | 51.1 | 6.72 | |
| CHARTOOL-7BParam.=7B2026.04 | 49.62 | — | |
| Deepeyes-V2Param.=7B2026.04 | 47.88 | — | |
| Qwen2.5-VL-72BParam.=72B2026.04 | 47.04 | — | |
| ThymeParam.=7B2026.04 | 46.55 | — | |
| START-RLModel size=7B2025.12 | 46.3 | — | |
| DeepeyesParam.=7B2026.04 | 45.55 | — | |
| Qwen2.5-VL-7B-Instruct + GRPO + KAWHITraining Method=GRPO + KAWHI2026.03 | 45.3 | — | |
| SketchVL-3BModel Scale=3B, Training Strategy=FinePO (Full Model)2026.01 | 44.15 | — | |
| Chart-R1Model size=7B2025.12 | 44 | — | |
| SketchVL-3BModel Scale=3B, Reward Mechanism=Random scores2026.01 | 43.94 | — | |
| Qwen2.5-VL-7B-Instruct + GRPOTraining Method=GRPO2026.03 | 43.7 | — | |
| SketchVL-3BModel Scale=3B, Training Strategy=Naive GRPO2026.01 | 43.69 | — | |
| Claude Sonnet 3.5Param.=–2026.04 | 43.58 | — | |
| SketchVL-3BModel Scale=3B, KL Action Regularization=Disabled2026.01 | 43.33 | — | |
| Qwen3-VL-8BParam.=8B2026.04 | 42.65 | — | |
| CHARTOOL-3BParam.=3B2026.04 | 42.54 | — | |
| InternVL3.5-8B-MPOModel Category=Open-source VLM2026.04 | 42 | 27.27 | |
| Qwen2.5-VL-7B-InstructModel Size=7B2026.03 | 41.9 | — | |
| START-SFTModel size=7B2025.12 | 41.8 | — | |
| Qwen2.5-VLModel size=7B2025.12 | 41.3 | — | |
| ChartReasonerModel size=7B2025.12 | 39.9 | — | |
| Qwen2.5-VL-7BParam.=7B2026.04 | 39.84 | — | |
| Qwen3-VL-4B-InstructModel Category=Open-source VLM2026.04 | 39.6 | 10.04 | |
| VLM-R12026.01 | 39.58 | — | |
| SketchVL-3BModel Scale=3B, Training Strategy=Cold Start only2026.01 | 38.45 | — | |
| START-RLModel size=3B2025.12 | 38.2 | — | |
| InternVL3-8BParam.=8B2026.04 | 37.78 | — | |
| Qwen2.5VL-3BModel Scale=3B2026.01 | 37.73 | — | |
| GPT-4oParam.=–2026.04 | 37.67 | — | |
| START-SFTModel size=3B2025.12 | 34.2 | — | |
| Qwen2.5-VLModel size=3B2025.12 | 32.9 | — | |
| Janus-pro-7BModel Category=Open-source VLM2026.04 | 32 | 18.25 | |
| Qwen2.5-VL-3BParam.=3B2026.04 | 30.28 | — | |
| ChartMOEParam.=8B2026.04 | 29.1 | — | |
| SketchVL-3BModel Scale=3B, Reasoning Method=Zero GRPO (No Sketch)2026.01 | 25.92 | — | |
| ECDModel size=7B2025.12 | 22.5 | — | |
| ECD Qwen2.5-VL-7BParam.=7B2026.04 | 22.5 | — | |
| SketchVL-3BModel Scale=3B, Training Strategy=SFT (No RL)2026.01 | 20.02 | — | |
| TinyChartParam.=3B2026.04 | 13.25 | — | |
| TinyChartModel size=3B2025.12 | 13.2 | — | |
| ChartGemmaParam.=3B2026.04 | 6.84 | — | |
| ChartGemmaModel size=3B2025.12 | 6.8 | — |