SketchAgentBench – evaluate LLMs on 3D spatial reasoning
AI summary
Editorial policyA benchmark evaluates LLMs on 3D spatial reasoning by scoring 11 models, 25 photos, and 275 runs based on their final 3D scenes generated from a single photo and prompt.
Key points
- •It evaluates the workflow from a single photo and prompt to code generation, rendering, inspection, and refinement.
- •The benchmark tests 11 models, 25 photos, and 275 runs.
- •Each model is scored based on its final 3D scene.
#Research & Papers#LLM / Model Trends#SketchAgentBench