AutoCAD Bench Introduced to Test Multimodal Agents in 3D Modeling
Markov Studios has introduced AutoCAD Bench, a benchmark designed to test the capabilities of multimodal agents in translating dimensioned reference drawings into correct, editable DWG files. This benchmark spans 50 tasks, covering both 2D drafting and 3D reconstruction. Each task provides agents with a prompt, a reference image, and a clean AutoCAD document, challenging them to execute visible mouse and keyboard actions to complete the task. The benchmark evaluates the agents' ability to perform tasks across four difficulty levels, with a focus on geometry, dimensions, annotations, and presentation. 
