Report
RoTA-Bench tests 3D spatial reasoning using 2D views
HuggingPapers says humans solve about 100% of its tasks, while GPT-5.6-Sol solves only about 17%.
TLDR
HuggingPapers describes RoTA-Bench as a new benchmark that asks models to align projections by rotating objects in 3D from 2D views. It says humans solve about 100% of the tasks, compared with about 17% for GPT-5.6-Sol.
Combined views
2.2K
2 Sources, first seen ago
