XR / Spatial Computing
The First Benchmark for Coding Agents in XR Says the Best One Passes Half the Tasks — and Its Code Is Not Where It Says
An unreviewed preprint from a 16-author team introduces GenXR-Bench, running held-out tests on consumer XR hardware with 96.5% agreement against blinded expert judgement. It calls itself a reproducible testbed. The GitHub URL printed in the paper is not a repository.