GenCAD — On Our Radar

Status: 🟡 Under observation
Why we're talking about it: GenCAD utilises autoregressive transformers and latent diffusion to transform 2D sketches into parametric CAD sequences, though the project remains in a fragile research state (source: GitHub).

What we know:
- The framework was accepted by TMLR in July 2025, with subsequent "Self-Repairing" iterations presented at ASME 2025 to address a 10% rate of infeasible B-reps (source: OpenReview.net, ASME.org).
- The official GitHub repository is currently hindered by 26+ open issues regarding broken Docker builds and missing 'pythonocc-core' dependencies (source: GitHub).

The unknowns:
- We don't know yet if a roadmap exists for a stable 1.0 release or a cloud-hosted version to bypass the current dependency hell.
- Public benchmarks comparing its output accuracy to 2026 industrial tools like Zoo.dev or Fusion 360 AI are currently missing.

This article will be updated when we have more data. Until then, proceed with caution. The high-quality outputs on the project site currently feel more like aspiration than accessible engineering.

Ship clean code,
Marcus.