benchgap
Vision & documents

BenchCAD Vision2Code (no tools) leaderboard

As of 2026-11-18, the highest measured score on BenchCAD Vision2Code (no tools) is 0.7% by Claude Sonnet 5.5. 5 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Claude Fable 5.20.8%estimated ± 0.1 pp, low confidence
2GPT-6 Astra0.8%estimated ± 0.1 pp, medium confidence
3Claude Sonnet 5.50.7%measured
4Claude Opus 5.50.7%measured
5Kimi K3.1 Pro0.7%measured
6Claude Fable 5.10.4%measured
7DeepSeek V4.1 Flash 10170.4%estimated ± 0.1 pp, medium confidence
8Claude Opus 50.4%measured
9Inkling 2 Small0.4%estimated ± 0.1 pp, low confidence
10GLM-5.3-Flash0.3%estimated ± 0.1 pp, low confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General