Vision-Intensive Perception on MME-Unify
13Spot DifferenceQwen2.5-VL-7B-Instruct
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Qwen2.5-VL-7B-InstructModel Scale=7B2026.01 | 13 | 32.7 | 48.8 | |
| R1-onevision-RLTraining Paradigm=RL2026.01 | 14 | 30.8 | 54.2 | |
| MM-EurekaQwen-7BModel Scale=7B2026.01 | 19 | 44.2 | 42.4 | |
| Thyme2026.01 | 21 | 46.2 | 57.2 | |
| LLaVA-OneVision-Qwen2-7BModel Scale=7B2026.01 | 27 | 46.2 | 50 | |
| Ours (SFT)Training Paradigm=SFT2026.01 | 27 | 34.6 | 55.2 | |
| Ours (RL)Training Paradigm=RL2026.01 | 28 | 38.5 | 55.1 | |
| Janus-Pro-7BModel Scale=7B2026.01 | 29 | 28.8 | — | |
| InternVL2.5-8BModel Scale=8B2026.01 | 30 | 32.7 | 49.6 |