CADBench: leaderboard
Recovering editable CAD programs from images or 3D observations is central to AI-assisted design, but progress is difficult to measure because existing evaluations are fragmented across datasets, modalities, and metrics. We introduce CADBench, a unified benchmark for multimodal CAD program generation. CADBench contains 18,000 evaluation samples spanning six benchmark families derived from DeepCAD,
Metric: IoU (imported). Source: benchmarklist.com. Saturation forecast: Around July 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 | 41.2 | #49 |
| 2 | Gemini 3.1 Pro (Preview) | 38.2 | #58 |
| 3 | Kimi K2.6 (Non-reasoning) | 22.4 | #104 (Kimi K2.6) |
| 4 | GPT-5.4 | 15.8 | #79 |
| 5 | Qwen 3.5 27B (Non-reasoning) | 1.5 | #187 (Qwen 3.5 27B) |
| 6 | Qwen 3.5 9B (Non-reasoning) | 0 | #379 (Qwen 3.5 9B) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=cadbench · How It Works · Data refreshed daily, snapshot 2026-10-20.