Chart Question Answering on MC v1 (test)
74.43AccuracyChartCynics
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| ChartCynicsBackbone=Qwen3-VL-8B, Configuration=SFT + GRPO (Full)2026.03 | 74.43 | 11.15 | 14.42 | |
| ChartCynicsBackbone=GPT-o4-mini, Configuration=Train-free2026.03 | 71.8 | 18.69 | 9.51 | |
| ChartCynicsBackbone=Gemini-2.5-Flash, Configuration=Train-free2026.03 | 71.67 | 13.67 | 14.67 | |
| Gemini-3.1-Procategory=Standard Multimodal Baseline2026.03 | 70.49 | 19.34 | 10.16 | |
| Gemini-2.5-FlashConfiguration=Base Model2026.03 | 67.21 | 20.98 | 11.8 | |
| ChartCynicsBackbone=Qwen3-VL-8B, Configuration=Train-free2026.03 | 60.66 | 24.26 | 15.08 | |
| GPT-o4-miniConfiguration=Base Model2026.03 | 55.08 | 34.43 | 10.49 | |
| Table-based QABackbone=Qwen3-VL-8B, category=Structural Mitigation Heuristic2026.03 | 49.18 | 29.51 | 21.31 | |
| Qwen3-VL-8BConfiguration=Base Model2026.03 | 45.57 | 40 | 14.43 | |
| RedrawnBackbone=Qwen3-VL-8B, category=Structural Mitigation Heuristic2026.03 | 44.26 | 35.41 | 20.33 | |
| ChartMoEcategory=Standard Multimodal Baseline2026.03 | 33.11 | 43.28 | 23.61 |