Microsoft Research· Yunfei Ge, Anbang Liu, Qineng Wang, Johnalbert Garnica, Zihan Wang, Reuben Tan, Jianfeng Gao, Ruohan Zhang, Yining Hong, Jiajun Wu, Manling Li·· 2026-08-13
MindTopo reveals VLMs’ spatial reasoning abilities
MindTopo reveals VLMs’ spatial reasoning abilities
AI summary
A path, a fence, a knot. MindTopo sets a new benchmark for testing how AI understands topological relationships and highlights new opportunities to strengthen spatial reasoning and planning.
Selection record
Threshold 60Official, first-handFirst 38Second 38
Not admittedSum of both 76 < twice the threshold 120
- Source tier
- Official, first-hand; this tier's threshold is 60
- Pre-filter
- passed:微软研究发布VLM拓扑推理基准MindTopo
A model scores each item twice, independently, against one written standard, out of 100. An item is admitted only when the two scores add up to twice the threshold. The threshold is set per source tier.
Source: Microsoft Research · microsoft.com