Vision Reasoning on Tunnel Vision
100ObjReID ScoreHuman
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Human2026.05 | 100 | 100 | 99 | |
| GPT-5.22026.05 | 71.8 | 68.4 | 52.9 | |
| Gemini 3.1 Pro2026.05 | 67.3 | 72 | 46.4 | |
| CAVE-7BParam=7B, Instruct=false2026.05 | 53.5 | 16.7 | 23.8 | |
| R1-OneVisionParam=7B, Instruct=false2026.05 | 52.6 | 8 | 16.7 | |
| CAVE-3BParam=3B, Instruct=false2026.05 | 50.2 | 12.9 | 21.1 | |
| Random2026.05 | 50 | 12.5 | 4 | |
| MiniCPM-V 4.5Param=8B, Instruct=false2026.05 | 50 | 11.8 | 17.8 | |
| CAVE Belief-OnlyParam=7B, Instruct=false2026.05 | 49.7 | 12.2 | 20.9 | |
| DeepEyesParam=7B, Instruct=false2026.05 | 49.3 | 13.3 | 19.1 | |
| Qwen3-VLParam=8B, Instruct=true2026.05 | 48.2 | 13.6 | 22.4 | |
| VLAA-ThinkerParam=7B, Instruct=false2026.05 | 46.9 | 10.9 | 22.2 | |
| ThinkLite-VLParam=7B, Instruct=false2026.05 | 46.9 | 10.2 | 22.7 | |
| InternVL3.5Param=8B, Instruct=true2026.05 | 45.1 | 15.3 | 20.4 | |
| Qwen2.5-VLParam=7B, Instruct=true2026.05 | 44.4 | 11.1 | 17.6 |