Chart Question Answering on CharXiv (test)
47.9AccuracyGPT-4o
Evaluation results
| Method | Links | |
|---|---|---|
| GPT-4oModel Scale=Large Scale2026.06 | 47.9 | |
| VTool-R1 (GRPO)Model Scale=Large Scale, Backbone=Qwen-2.5VL-7B2026.06 | 39.2 | |
| Qwen-2.5VL-7BModel Scale=Large Scale2026.06 | 31 | |
| IAPOModel Scale=Small Scale, Backbone=Qwen-2.5VL-3B2026.06 | 26.1 | |
| ToRLModel Scale=Small Scale, Backbone=Qwen-2.5VL-3B2026.06 | 24.1 | |
| TRMModel Scale=Small Scale, Backbone=Qwen-2.5VL-3B2026.06 | 23.6 | |
| VTool-R1 (GRPO)Model Scale=Small Scale, Backbone=Qwen-2.5VL-3B2026.06 | 23.1 | |
| Qwen-2.5VL-3BModel Scale=Small Scale2026.06 | 18.9 |