Vision Reasoning on Tunnel Vision, BabyVision, and TRACER-Bench
98.3Overall ScoreHuman
Evaluation Results
| Method | Links | |
|---|---|---|
| Human2026.05 | 98.3 | |
| GPT-5.22026.05 | 61.8 | |
| Gemini 3.1 Pro2026.05 | 56.4 | |
| CAVE-7BParam=7B, Instruct=false2026.05 | 33.3 | |
| CAVE-3BParam=3B, Instruct=false2026.05 | 30.5 | |
| Qwen3-VLParam=8B, Instruct=true2026.05 | 30.1 | |
| CAVE Belief-OnlyParam=7B, Instruct=false2026.05 | 29.6 | |
| InternVL3.5Param=8B, Instruct=true2026.05 | 29.2 | |
| MiniCPM-V 4.5Param=8B, Instruct=false2026.05 | 29.2 | |
| DeepEyesParam=7B, Instruct=false2026.05 | 28.5 | |
| ThinkLite-VLParam=7B, Instruct=false2026.05 | 27.4 | |
| VLAA-ThinkerParam=7B, Instruct=false2026.05 | 27 | |
| R1-OneVisionParam=7B, Instruct=false2026.05 | 26 | |
| Qwen2.5-VLParam=7B, Instruct=true2026.05 | 25.9 | |
| Random2026.05 | 24.1 |