OCR VQA on ChartQA (test)
87.92AccuracyMIRROR (ours)
Evaluation results
| Method | Links | |
|---|---|---|
| MIRROR (ours)Param Size=7B2026.02 | 87.92 | |
| MIRROR (w/o tool)Param Size=7B2026.02 | 86.56 | |
| Qwen2.5-VL-7BParam Size=7B2026.02 | 86.08 | |
| FineViT-VLLanguage Backbone=Qwen3 1.7B LLM, Evaluation Framework=VLMEvalKit [69]2026.03 | 85.6 | |
| SAIL-VLModel Size=8B2025.01 | 84.6 | |
| DeepSeekVL-2Model Size=8B2025.01 | 84.6 | |
| Qwen2.5-VL-3BParam Size=3B2026.02 | 83.2 | |
| Qwen2-VLModel Size=8B2025.01 | 81.6 | |
| DeepSeekVL-2Model Size=2B2025.01 | 81.2 | |
| SAIL-VLModel Size=2B2025.01 | 81 | |
| Intern3.5-VLLanguage Backbone=Qwen3 1.7B LLM, Evaluation Framework=VLMEvalKit [69]2026.03 | 80.76 | |
| LLaVA-OneVisionParam Size=7B2026.02 | 80 | |
| InternVL3Param Size=8B2026.02 | 79.64 | |
| Qwen3-VLLanguage Backbone=Qwen3 1.7B LLM, Evaluation Framework=VLMEvalKit [69]2026.03 | 77.72 | |
| InternVL2.5-MPOModel Size=8B2025.01 | 77.6 | |
| Aquila-VLLanguage Backbone=Qwen2.5 1.5B LLM, Evaluation Framework=VLMEvalKit [69]2026.03 | 76.52 | |
| InternVL3Param Size=2B2026.02 | 76.08 | |
| InternVL2.5-MPOModel Size=2B2025.01 | 70.9 | |
| Qwen2-VLModel Size=2B2025.01 | 70.6 | |
| Emu3Param.=8B2026.02 | 68.6 | |
| HIVEBackbone=Huginn, Param.=4B, Train Data=6.5M, Recurrence Level (r)=322026.02 | 67 | |
| Gemma3-4B-PTBackbone=Gemma3, Param.=4B, Train Data=4T*2026.02 | 63.6 |