Cogira PSR-Bench (Procedural Spatial Reasoning Benchmark) generates infinite, non-memorisable visual reasoning tests for evaluating multimodal LLMs. Each test places shapes on a canvas using chained geometric rules — rotation, scaling, reflection, offset — where every shape depends on previously placed ones. A single early error cascades through the entire layout.
Configure a spatial reasoning test (or just go with the defaults), then hit Generate.