CADEngBench - Functional Editing (L2-E): leaderboard
The functional-editing task of CADEngBench (MBZUAI, 2026): the model receives source CAD, CadQuery or Fusion 360 history, and an edit request, and an edit passes when the requested change is measured and protected non-target geometry is preserved. Percentage of 300 edit tasks passed; higher is better.
Metric: Edit pass rate (%; share of 300 functional-edit tasks on supplied source CAD, native CadQuery or a replayed Fusion 360 JSON patch, where the requested change is measured and protected non-target geometry is preserved; one response per task). Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Preview) (Minimal) | 72.3 | #144 (Gemini 3 Flash (Preview)) |
| 2 | GPT-5.2 (2025-12-11) (Non-reasoning) | 70.3 | |
| 3 | Kimi K2.5 (Non-reasoning) | 67 | #256 (Kimi K2.5) |
| 4 | Claude Sonnet 4.5 | 66.7 | #250 |
| 5 | Qwen 3.5 35B A3B (Non-reasoning) | 61.7 | #482 (Qwen 3.5 35B A3B) |
| 6 | Llama 4 Maverick | 60.7 | #1099 |
| 7 | GLM-4.6V (Non-reasoning) | 59.3 | #701 (GLM-4.6V) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=cadengbench-functional-editing-l2-e · How It Works · Data refreshed daily, snapshot 2026-11-13.