Visual Navigation on TIR-Bench Maze
65AccuracyQwen3-VL + V-ABS
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen3-VL + V-ABSBackbone=Qwen3-VL-8B, Strategy=V-ABS2026.05 | 65 | |
| Intern-VL3-8B + V-ABSBackbone=Intern-VL3-8B, Strategy=V-ABS2026.05 | 56.7 | |
| Qwen2.5-VL + V-ABSBackbone=Qwen2.5-VL-7B, Strategy=V-ABS2026.05 | 46.7 | |
| Intern-VL3-8BBackbone=Intern-VL3-8B2026.05 | 26.7 | |
| GPT-4o + V-ABSBackbone=GPT-4o, Strategy=V-ABS2026.05 | 23.3 | |
| GPT-4o + VisuoThinkBackbone=GPT-4o, Strategy=VisuoThink2026.05 | 20.1 | |
| Qwen3-VL-8BBackbone=Qwen3-VL-8B2026.05 | 17.5 | |
| GPT-4oBackbone=GPT-4o2026.05 | 17.5 | |
| Qwen2.5-VL-7BBackbone=Qwen2.5-VL-7B2026.05 | 10.7 |