gNucleus AI publishes Parametric CAD Bench v3 for AI agents that author parametric CAD
Benchmark cadbench.aiPaper arxiv.org
gNucleus AI publishes the paper for Parametric CAD Bench v3, which grades AI agents on parametric design intent, modelling-history preservation and feature-level editability, not only on whether a generated 3D model looks right. It has 100 real-world-inspired tasks in three workflows: text-to-CAD creation, text-to-CAD create-and-edit, and image-to-CAD from engineering drawings. The paper reports a snapshot of 13 model-agent systems as of 24 September 2026; according to a company author, Claude Code with Opus 5.5 leads with an overall 61.03% and 84.66% on image-to-CAD. The benchmark, leaderboard and trajectories were released in September.
Share on LinkedIn, X or Bluesky, or .
Know related work that belongs here? Suggest a link
Something wrong on this page? Tell the editor
More news
All news- ReleaseUniversalAGI releases SUV-PT, an SUV aerodynamics model with 1.1% drag error on DrivAerML
- EventAutoCFD5 uses DrivAerML as the common training set for its AI/ML comparison of road-car surrogates
- ReleaseLuminary's SHIFT-NVH predicts road-induced body vibration from part gauges, material and speed
- ReleaseXplore's RWE-Bench grades AI agents with physics solvers on CAD, PCB and CFD tasks