BenchCAD 1.0 - Code Edit: leaderboard

Given a CadQuery program and a plain-language edit instruction, the model returns the modified program, scored over 748 edit pairs by how much of the IoU gap between the original and target solids its edit closes. Unlike BenchCAD 1.0 Vision2Code it starts from existing code rather than images.

Metric: Accuracy (0 to 1, share of the original-to-target IoU gap the edit closes) on 748 CadQuery edit pairs. Source: benchcad.com. Saturation forecast: Around February 2027. 5 models tracked. Updated 16 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Claude Opus 4.7Thinking0.853#78
2Gemini 3.1 Pro (Preview)Thinking0.837#93
3O30.708#228
4GPT-4o0.615#744
5Nemotron 3 Super 120B A12BNon-reasoning0.608

No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.

Interactive version: theaggregate.ai/benchmark?slug=benchcad-1-0-code-edit · How It Works · Data refreshed daily, snapshot 2026-11-16.