Multimodal Chart Analysis on MCDR-Bench
81.21BG ScoreGemini-2.5-Pro
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Gemini-2.5-Pro2026.03 | 81.21 | 87.31 | 91.44 | 93.78 | 92.98 | 89.29 | 89.34 | |
| Claude-3.7 Sonnet2026.03 | 68.78 | 57.27 | 89.45 | 85.02 | 86.97 | 74.96 | 77.08 | |
| Qwen2.5-VL-7B-Instruct w/ PRPO ThinkOptimization Strategy=PRPO, Prompt Setting=chain-of-thought reasoning (Think)2026.03 | 62.9 | 65.23 | 88.87 | 80.91 | 87.21 | 76.26 | 76.89 | |
| Qwen2.5-VL-7B-Instruct w/ PRPOOptimization Strategy=PRPO, Prompt Setting=direct answer generation2026.03 | 50.65 | 61.38 | 81.78 | 72.83 | 84.01 | 69.62 | 69.9 | |
| Qwen2.5-VL-7B-Instruct w/ GRPO ThinkOptimization Strategy=GRPO, Prompt Setting=chain-of-thought reasoning (Think)2026.03 | 43.13 | 48.77 | 77.75 | 70.28 | 81.02 | 63 | 63.99 | |
| Qwen2.5-VL-7B-Instruct w/ GRPOOptimization Strategy=GRPO, Prompt Setting=direct answer generation2026.03 | 41.24 | 51.69 | 75.38 | 66.14 | 77.4 | 61.71 | 62.26 | |
| GPT-4.12026.03 | 34.68 | 19.5 | 72.25 | 77.11 | 79.22 | 50.87 | 55.61 | |
| Llama-3.2-11B-Vision2026.03 | 28.06 | 33.02 | 36.02 | 36.41 | 39.44 | 34.4 | 34.56 | |
| GPT-4o2026.03 | 27.17 | 21.89 | 41.01 | 47.45 | 60 | 35.83 | 38.89 | |
| Qwen2.5-VL-7B-InstructPrompt Setting=direct answer generation2026.03 | 23.35 | 39.43 | 51.04 | 37.59 | 45.63 | 40.01 | 39.51 | |
| InternVL2.5-8B2026.03 | 21.09 | 20.77 | 37.27 | 32.67 | 47.12 | 30.64 | 31.59 | |
| LLaVA-Next-7B2026.03 | 10.73 | 7.81 | 23.08 | 20.66 | 27.29 | 16.96 | 17.76 | |
| ChartGemma-3B2026.03 | 7.53 | 5.02 | 7.51 | 9.25 | 14.29 | 8.14 | 8.29 | |
| ChartInstruct-LLama2-7B2026.03 | 4.9 | 3.38 | 5.84 | 6.69 | 5.54 | 5.09 | 5.24 | |
| Qwen2.5-VL-3B-Instruct2026.03 | 1.88 | 0.81 | 3.06 | 10.83 | 12.58 | 4.96 | 5.69 |