CADEngBench - Executable CAD (L0): leaderboard
CADEngBench (MBZUAI, 2026) evaluates multimodal models as CAD engineers. In the zero-to-CAD task a model turns an engineering brief, reference views, parameters and requirements into an executable CadQuery program; L0 is the percentage of 300 parts whose program runs, builds a valid B-Rep solid and survives STEP export and re-import, higher is better.
Metric: Executable CAD rate (%; share of 300 parametric part tasks (159 BenchCAD, 141 Fusion 360) whose brief, reference views, parameters and requirements the model turns into an executable CadQuery program; one response per task, no retries whose program executes, builds a valid B-Rep solid and survives STEP export and re-import). Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 58 | #250 |
| 2 | Kimi K2.5 (Non-reasoning) | 53.3 | #256 (Kimi K2.5) |
| 3 | Gemini 3 Flash (Preview) (Minimal) | 52.7 | #144 (Gemini 3 Flash (Preview)) |
| 4 | Llama 4 Maverick | 48 | #1099 |
| 5 | GLM-4.6V (Non-reasoning) | 47.3 | #701 (GLM-4.6V) |
| 6 | GPT-5.2 (2025-12-11) (Non-reasoning) | 41.3 | |
| 7 | Qwen 3.5 35B A3B (Non-reasoning) | 13.3 | #482 (Qwen 3.5 35B A3B) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=cadengbench-executable-cad-l0 · How It Works · Data refreshed daily, snapshot 2026-11-13.