SpatialBench, introduced in a new paper, offers a suite of 146 verifiable problems drawn from real spatial biology workflows across five platforms (MERFISH, Seeker, Visium, Xenium, ATLASxomics DBIT‑seq) to evaluate agent‑based analysis tools. The benchmark reproduces decision points scientists face—QC, normalization, clustering, cell typing and spatial analysis—and measures verifiable biological outcomes rather than superficial outputs. Authors argue SpatialBench will accelerate robust deployment of agents in labs by providing realistic, reproducible metrics.