Chart-to-Code on ChartMimic Customized
95Execution RateVinciCoder-8B-SFT
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| VinciCoder-8B-SFTReinforcement Learning Strategy=RL (Visual-ERM)2026.03 | 95 | 60.9 | 82.6 | 71.7 | |
| Qwen3-VL-8B-InstructReinforcement Learning Strategy=RL (DINO-based Reward)2026.03 | 91.7 | 71.2 | 80.3 | 75.8 | |
| Qwen3-VL-8B-InstructReinforcement Learning Strategy=RL (Visual-ERM)2026.03 | 91.5 | 71.8 | 81.1 | 76.5 | |
| VinciCoder-8B-SFTReinforcement Learning Strategy=RL (DINO-based Reward)2026.03 | 91.5 | 57 | 77.5 | 67.3 | |
| ChartMaster-7BModel=ChartMaster-7B2026.03 | 88.8 | 59.5 | 74.2 | 66.8 | |
| Qwen3-VL-8B-InstructModel=Qwen3-VL-8B-Instruct2026.03 | 86.8 | 66.3 | 76.8 | 71.6 | |
| VinciCoder-8B-SFTModel=VinciCoder-8B-SFT2026.03 | 86.2 | 52.3 | 71.5 | 61.9 | |
| JanusCoderV-8BModel=JanusCoderV-8B2026.03 | 80.7 | 66.7 | 74.2 | 70.4 | |
| InternVL3.5-8BModel=InternVL3.5-8B2026.03 | 79.7 | 53 | 63.3 | 58.2 |