neuralCAD-Edit: leaderboard

neuralCAD-Edit (Autodesk AI Lab, 2026) is an expert benchmark for multimodal-instructed 3D CAD editing: designers record edit requests while talking, pointing and drawing, and models edit the CAD model by writing CadQuery scripts in a common harness. This board is the expert acceptance rate in percent, the share of edits whose median ratings from five CAD experts reach the acceptable threshold for both instruction following and quality; higher is better.

Metric: Expert acceptance rate (from 0-1, times 100): share of edits whose median instruction-following and quality ratings from five CAD experts both reach the acceptable threshold (5 of 7), on the 192 neuralCAD-Edit requests (expert CAD designers' video, speech, drawing and text edit requests on Autodesk Fusion models), the model editing the input STEP model by writing and re-running CadQuery scripts in the authors' harness with visual inspection; higher is better. Source: arxiv.org. Saturation forecast: Around July 2028. 3 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-5.225#190
2Gemini 3 Pro10#146
3Claude Sonnet 4.55#267

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=neuralcad-edit · How It Works · Data refreshed daily, snapshot 2026-11-22.