We introduce GGBench, a geometric generative reasoning benchmark purpose-built for unified multimodal models (UMMs). Unlike prior evaluations that treat discriminative understanding and unconstrained image generation separately, GGBench diagnoses whether a model can fuse language comprehension with precise visual construction. Geometric construction serves as an ideal testbed, revealing how well a system can actively reason and synthesize structured solutions across modalities.

Unified multimodal models (UMMs) herald a shift from passive perception toward proactive, cross-modal generation. However, current benchmarks rarely stress-test whether these systems can integrate reasoning with controlled synthesis. Most evaluations remain disjoint—either probing language understanding or measuring image fidelity in isolation. As a result, we still lack a principled way to measure generative reasoning.

GGBench closes this gap by framing geometric construction as a rigorous reasoning task. A model must parse natural-language specifications, plan a construction, and render accurate intermediate artifacts. This workflow surfaces fine-grained failure modes in alignment, consistency, and controllability—dimensions that traditional captioning or free-form generation overlook. By standardizing data, protocol, and diagnostics, GGBench offers researchers a reproducible lens on how UMMs evolve from understanding to deliberate problem solving.

We curate GGbench Benchmark, a comprehensive benchmark providing a standardized taxonomy and an evaluation protocol, enabling consistent and category-wise assessment beyond surface-level metrics.