BenchCAD 1.0 - Vision QA: leaderboard

From four rendered views of a CAD part, the model answers numeric geometry questions (ratios within 5%, integers exactly) across 2,400 questions; the score is the share answered correctly. It asks the same questions as BenchCAD 1.0 Code QA, which gives the CadQuery source instead of images.

Metric: Total accuracy (0 to 1) on 2,400 numeric CAD questions from multi-view renders. Source: benchcad.com. Saturation forecast: Around September 2027. 5 models tracked. Updated 16 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 3.1 Pro (Preview)Non-reasoning0.587#93
2Claude Opus 4.7Thinking0.53#78
3GPT-5.30.513#197
4GPT-4o0.464#744
5O30.327#228

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-vision-qa · How It Works · Data refreshed daily, snapshot 2026-11-16.