Chart understanding on ChartX
83.51GPT ScoreGPT-5
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-5Model Size=-, # Training Data=-2026.05 | 83.51 | |
| Gemini-2.5-ProModel Size=-, # Training Data=-2026.05 | 74.22 | |
| Qwen2.5-VL + ChartCF (Ours)Model Size=7B, # Training Data=4K pairs (DPO)2026.05 | 73.18 | |
| Qwen3-VL + ChartCF (Ours)Model Size=8B, # Training Data=4K pairs (DPO)2026.05 | 72.14 | |
| Qwen3-VL + ECDModel Size=8B, # Training Data=300K (SFT)2026.05 | 71.61 | |
| Qwen2.5-VL + ECDModel Size=7B, # Training Data=300K (SFT)2026.05 | 70.83 | |
| InternVL3.5 + ECDModel Size=8B, # Training Data=300K (SFT)2026.05 | 70.31 | |
| InternVL3.5Model Size=8B, # Training Data=-2026.05 | 68.49 | |
| InternVL3.5 + ChartCF (Ours)Model Size=8B, # Training Data=4K pairs (DPO)2026.05 | 68.4 | |
| Qwen2.5-VLModel Size=7B, # Training Data=-2026.05 | 67.8 | |
| Qwen3-VLModel Size=8B, # Training Data=-2026.05 | 67.8 | |
| GPT-4oModel Size=-, # Training Data=-2026.05 | 58.33 | |
| ReachQA (InternVL2-8B)Model Size=8B, # Training Data=20K (SFT)2026.05 | 45.38 | |
| GPT-4o miniModel Size=-, # Training Data=-2026.05 | 44.36 | |
| Claude-3.5-SonnetModel Size=-, # Training Data=-2026.05 | 42.71 | |
| ChartGemmaModel Size=3B, # Training Data=123K (SFT)2026.05 | 35.15 | |
| ChartSpec + 4kParams=7B2026.02 | 3.52 | |
| ChartSpec + 3kParams=7B2026.02 | 3.38 | |
| GPT-4V2026.02 | 2.63 | |
| Qwen2.5-VL-7B(backbone)Params=7.0B2026.02 | 2.23 | |
| ChartCoderParams=7.0B2026.02 | 2.09 | |
| TinyChartParams=3B2026.02 | 1.89 | |
| InternVL2-Llama3-76BParams=76.0B2026.02 | 1.74 | |
| InternVL2-26BParams=26.0B2026.02 | 1.7 | |
| Qwen2-VL-72BParams=72.0B2026.02 | 1.69 | |
| MiniCPM-Llama3-V2.5Params=8.4B2026.02 | 1.66 | |
| InternVL2-8BParams=8.1B2026.02 | 1.63 | |
| ChartVLM-LParams=14.3B2026.02 | 1.58 | |
| Qwen2-VL-7BParams=7.0B2026.02 | 1.5 | |
| ChartLlamaParams=13B2026.02 | 0.94 | |
| ChartAssistantParams=13B2026.02 | 0.82 |