Document Question Answering on CharXiv RQ
68.6AccuracyGPT-5 mini
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-5 miniVersion=high2026.05 | 68.6 | |
| RLR³Training Source=OpenMMR2026.05 | 62.7 | |
| RLVRTraining Source=ViRL2026.05 | 62.4 | |
| RLR³Training Source=DeepVision2026.05 | 61.8 | |
| RLR³Training Source=ViRL2026.05 | 61.4 | |
| RLVRTraining Source=DeepVision2026.05 | 61.2 | |
| RLVRTraining Source=OpenMMR2026.05 | 58.4 | |
| Official thinkingSource=Qwen3-VL2026.05 | 56.6 | |
| Base instruct2026.05 | 53.4 | |
| Official instructSource=Qwen3-VL2026.05 | 48.9 | |
| GPT-5 miniVersion=minimal2026.05 | 48.9 |