Mathematical & Geometric Reasoning on MMK12
86.4Accuracy@8Qwen2.5-VL-32B + DAPO
Evaluation results
| Method | Links | |
|---|---|---|
| Qwen2.5-VL-32B + DAPOModel Scale=32B, Algorithm=DAPO2025.10 | 86.4 | |
| Qwen2.5-VL-32B + VPPOModel Scale=32B, Algorithm=VPPO2025.10 | 86.3 | |
| Qwen2.5-VL-7B + VPPOModel Scale=7B, Algorithm=VPPO2025.10 | 82.8 | |
| Qwen2.5-VL-7B + DAPOModel Scale=7B, Algorithm=DAPO2025.10 | 82.1 | |
| Qwen2.5-VL-32B + GRPOModel Scale=32B, Algorithm=GRPO2025.10 | 80.7 | |
| PAPO-D-7BModel Scale=7B, Training Strategy=Pure RL2025.10 | 80.6 | |
| MM-Eureka-32BModel Scale=32B, Prompting Source=Official author-provided2025.10 | 73.4 | |
| Qwen2.5-VL-7B + GRPOModel Scale=7B, Algorithm=GRPO2025.10 | 72.3 | |
| R1-ShareVL-7BModel Scale=7B, Training Strategy=Pure RL, Prompting Source=Official author-provided2025.10 | 70.9 | |
| Qwen2.5-VL-32BModel Scale=32B2025.10 | 68.8 | |
| VL-Rethinker-7BModel Scale=7B, Training Strategy=Pure RL, Prompting Source=Official author-provided2025.10 | 68.3 | |
| MM-Eureka-7BModel Scale=7B, Training Strategy=Pure RL, Prompting Source=Official author-provided2025.10 | 67.5 | |
| ThinkLite-7BModel Scale=7B, Training Strategy=Pure RL, Prompting Source=Official author-provided2025.10 | 62.6 | |
| NoisyRollout-32BModel Scale=32B, Prompting Source=Official author-provided, Note=Trained using the training set of Geo3k2025.10 | 60.2 | |
| NoisyRollout-7BModel Scale=7B, Training Strategy=Pure RL, Prompting Source=Official author-provided, Note=Trained using the training set of Geo3k2025.10 | 50 | |
| Qwen2.5-VL-7BModel Scale=7B2025.10 | 42.5 |