Perception-intensive Reasoning on MME-RealWorld
59.1AccuracyOpenVLThinker
Evaluation Results
| Method | Links | |
|---|---|---|
| OpenVLThinker#Param=7B2026.04 | 59.1 | |
| Perceval#Param=7B2026.04 | 50 | |
| Vision-R1#Param=7B2026.04 | 49.6 | |
| Qwen2.5-VL + GRPO#Param=7B2026.04 | 49.5 | |
| VLAA-Thinker#Param=7B2026.04 | 48 | |
| Perceval#Param=3B2026.04 | 47.6 | |
| Qwen2.5-VL + GRPO#Param=3B2026.04 | 46.8 | |
| DeepEyes#Param=7B, Tool Capability=true2026.04 | 46.5 | |
| MM-Eureka#Param=7B2026.04 | 46.4 | |
| Pixel-Reasoner#Param=7B, Tool Capability=true2026.04 | 43.5 | |
| Qwen2.5-VL#Param=7B2026.04 | 43 | |
| VLM-R1#Param=3B2026.04 | 42.3 | |
| Jigsaw-R1#Param=3B2026.04 | 42.2 | |
| Qwen2.5-VL#Param=3B2026.04 | 41.7 | |
| R1-VL#Param=7B2026.04 | 40.5 | |
| VL-Rethinker#Param=7B2026.04 | 38.8 | |
| Perception-R1#Param=3B2026.04 | 37.5 | |
| LMM-R1#Param=3B2026.04 | 35.8 | |
| R1-VL#Param=2B2026.04 | 32.4 |