| Benchmark | Model | Rank | Results |
|---|---|---|---|
| chart-question-answering-on-charxiv-dq | Qwen2.5-VL-7B (CRM RL, after RL) | – | Accuracy: 77.3 |
| mathematical-reasoning-on-mathverse | CRM-RL-8B (Ours, soft-reward) | – | Accuracy: 65.8 |
| multimodal-reasoning-on-mathvision | CRM-RL-8B (Ours, soft-reward) | – | Accuracy: 57.2 |
| multimodal-reasoning-on-mathvista | CRM-RL-8B (Ours, soft-reward) | – | Accuracy: 79.2 |