Chart Understanding on ChartQA (test)
90.8AccuracyClaude-3.5 Sonnet
Evaluation Results
| Method | Links | |
|---|---|---|
| Claude-3.5 Sonnet2025.02 | 90.8 | |
| Claude-3.5-Sonnet2024.12 | 90.8 | |
| iVGR-Qwen3-VL-32BBackbone=Qwen3-VL, Model Size=32B, Training Method=iVGR2026.05 | 90.4 | |
| Qwen2.5-VL-72BModel Size=72B2025.02 | 89.5 | |
| OptMerge2025.05 | 88.92 | |
| iVGR-Qwen2.5-VL-7BBackbone=Qwen2.5-VL, Model Size=7B, Training Method=iVGR2026.05 | 88.5 | |
| InternVL2-76BModel Scale=76B2024.12 | 88.4 | |
| InternVL2.5-78BModel Size=78B2025.02 | 88.3 | |
| Qwen2-VL-72BModel Scale=72B2024.12 | 88.3 | |
| iVGR-Qwen3-VL-8BBackbone=Qwen3-VL, Model Size=8B, Training Method=iVGR2026.05 | 87.6 | |
| AndesVLNumber of Parameters=2.4B2025.12 | 87.4 | |
| Qwen2.5-VL-7BModel Size=7B2025.02 | 87.3 | |
| Gemini-1.5 Pro2024.10 | 87.2 | |
| Gemini 1.5 Pro2025.02 | 87.2 | |
| Gemini-1.5-Pro2024.12 | 87.2 | |
| Individual OCR2025.05 | 87 | |
| Individual Chart2025.05 | 86.96 | |
| Individual Geometry2025.05 | 86.76 | |
| GPT-4o2025.02 | 86.7 | |
| Qwen2.5-VL-7B + Cont. Rewardreward_type=Continuous2025.11 | 86.7 | |
| InternVL3.5-8BParameter Scale=8B2026.05 | 86.7 | |
| Individual VQA2025.05 | 86.68 | |
| ARIA2024.10 | 86.4 | |
| Qwen2.5-VL-7BBackbone=Qwen2.5-VL, Model Size=7B, Training Method=Baseline2026.05 | 86.4 | |
| InternVL3.5-4BParameter Scale=4B2026.05 | 86.4 | |
| MAmmoTH-VL-8BModel Scale=8B2024.12 | 86.2 | |
| Molmo2-4BParameter Scale=4B2026.05 | 86.1 | |
| PerceptionLM-8BParameter Scale=8B2026.05 | 86 | |
| Molmo2-8BParameter Scale=8B2026.05 | 86 | |
| MAmmoTH-VL-8BModel Scale=8B, Single-image variant (SI)=true2024.12 | 85.9 | |
| Individual Grounding2025.05 | 85.88 | |
| GPT-4o2024.10 | 85.7 | |
| GPT-4o2024.12 | 85.7 | |
| HyperVL ViTLNumber of Parameters=2.0B2025.12 | 85.7 | |
| Gemini-1.5 Flash2024.10 | 85.4 | |
| SAIL-VL2Number of Parameters=2.7B2025.12 | 85.4 | |
| Zamba2-VL-7BParameter Scale=7B2026.05 | 85.3 | |
| PerceptionLM-3BParameter Scale=3B2026.05 | 85.1 | |
| Qwen3-VL-32BBackbone=Qwen3-VL, Model Size=32B, Training Method=Baseline2026.05 | 85 | |
| LLaVA-OV-72BModel Scale=72B, Single-image variant (SI)=true2024.12 | 84.9 | |
| SAIL-VL1.5Number of Parameters=2.5B2025.12 | 84.8 | |
| Qwen2.5-VL-7B + Discrete Rewardreward_type=Discrete2025.11 | 84.62 | |
| Vision-Zeroexternal supervision=true2025.11 | 84.24 | |
| Molmo-7B-DModel Scale=7B2024.12 | 84.1 | |
| Qwen2.5-VL-3BModel Size=3B2025.02 | 84 | |
| InternVL2.5-4BComplexity=O(n)2025.12 | 84 | |
| Qwen2.5VL-3BComplexity=O(n)2025.12 | 84 | |
| Qwen2.5-VL-7B (Baseline)2025.11 | 84 | |
| HyperVLNumber of Parameters=1.8B2025.12 | 83.8 | |
| LLaVA-OV-72BModel Scale=72B2024.12 | 83.7 | |
| Qwen2.5-VL-32B-Instruct2025.05 | 83.64 | |
| Llama3.2-11B2024.10 | 83.4 | |
| Llama-3.2-11B-Vision-InsModel Scale=11B2024.12 | 83.4 | |
| InternVL-2-8BModel Scale=8B2024.12 | 83.3 | |
| Qwen3-VL-8BBackbone=Qwen3-VL, Model Size=8B, Training Method=Baseline2026.05 | 83.2 | |
| Qwen2-VL-7B-InsModel Scale=7B2024.12 | 83 | |
| Qwen2-VL-7BParameters=7B2026.06 | 83 | |
| Qwen3-VL-8BParameter Scale=8B2026.05 | 82.8 | |
| MiniCPM-V-2.6-7BModel Scale=7B2024.12 | 82.4 | |
| INXComp-2.5-7BModel Scale=2.5B2024.12 | 82.2 | |
| InfiniteVL-4BComplexity=O(1)2025.12 | 82 | |
| Pixtral-12B2024.10 | 81.8 | |
| Phi-3.5-Vision-4BComplexity=O(n)2025.12 | 81.8 | |
| Qwen3-VL-4BParameter Scale=4B2026.05 | 81.8 | |
| InternVL3.5-2BParameter Scale=2B2026.05 | 81.6 | |
| Qwen3VL-8BParameters=8B2026.06 | 81.2 | |
| PreciseDoc-Reasoner2026.06 | 81.2 | |
| InternVL3.5Number of Parameters=2.3B2025.12 | 80.7 | |
| InternVL3Number of Parameters=2.1B2025.12 | 80.2 | |
| LLaVA-OV-7BModel Scale=7B2024.12 | 80 | |
| Zamba2-VL-2.7BParameter Scale=2.7B2026.05 | 79.6 | |
| Ovis2Number of Parameters=2.5B2025.12 | 79.5 | |
| PerceptionLM-1BParameter Scale=1B2026.05 | 79.2 | |
| LLaVA-OV-7BModel Scale=7B, Single-image variant (SI)=true2024.12 | 78.8 | |
| Qwen3-VL-2BParameter Scale=2B2026.05 | 78.7 | |
| GPT-4V2024.10 | 78.4 | |
| InternVL3.5-1BParameter Scale=1B2026.05 | 78 | |
| Qwen3-VLNumber of Parameters=2.1B2025.12 | 77.6 | |
| Zamba2-VL-1.2BParameter Scale=1.2B2026.05 | 77.6 | |
| Iso-CMethod Class=Merging2025.05 | 76.96 | |
| TA w/ DAREMethod Class=Merging2025.05 | 76.48 | |
| TSV MergingMethod Class=Merging2025.05 | 76.44 | |
| OptMergeMethod Class=Merging2025.05 | 76.24 | |
| TIES w/ DAREMethod Class=Merging2025.05 | 76.04 | |
| TIES MergingMethod Class=Merging2025.05 | 75.48 | |
| Task ArithmeticMethod Class=Merging2025.05 | 75.28 | |
| WUDI MergingMethod Class=Merging2025.05 | 74.36 | |
| Qwen2-VLNumber of Parameters=2.2B2025.12 | 73.5 | |
| Cambrian-1-8BModel Scale=8B2024.12 | 73.3 | |
| TA w/ DAREProtocol=full fine-tuning2025.05 | 73.08 | |
| Task ArithmeticProtocol=full fine-tuning2025.05 | 72.88 | |
| TIES w/ DAREProtocol=full fine-tuning2025.05 | 72.72 | |
| TIES MergingProtocol=full fine-tuning2025.05 | 72.28 | |
| Weight AverageProtocol=full fine-tuning2025.05 | 71.64 | |
| Qwen3.5-9BParameters=9B2026.06 | 70.7 | |
| TSV MergingProtocol=full fine-tuning2025.05 | 70.56 | |
| Iso-CProtocol=full fine-tuning2025.05 | 70.56 | |
| Mixture TrainingProtocol=full fine-tuning2025.05 | 70.32 | |
| DocOwl-1.5-Chat2026.06 | 70.2 | |
| Qwen2-VL-InstructMethod Class=Baseline2025.05 | 70.04 |