Pen 2.1: Runs faster and finds more errors
Reviews come back more than twice as fast and Pen catches more mistakes. A new model, GPT-6 Astra, and a sharper review method.
Also on YouTube.
Pen 2.1 is live. Pen helps you release better drawings and avoid nasty back-and-forth between design and manufacturing.
In this version, we switched to GPT-6 Astra. Astra is the best frontier model for 3D understanding I have seen. An engineering drawing is a 3D part described in 2D views, so that capability helps us a lot. It shows in our own benchmarks, and it shows on BenchCAD, a 3D CAD benchmark. There, Astra is a clear step up.
If you play with it, you’ll probably get to the same conclusion. Astra is also fast. That’s why reviews now come back more than twice as fast.
Independent of the model switch, we also improved the way we review your drawings. You’ll see the strongest improvement in GD&T as well as in understanding manufacturing drawings for machining and sheet metal bending.
On our current main test set, Pen now finds 83% of all errors, up from 69%. That also tells you it’s not perfect. But every error caught early is one that doesn’t land on a colleague’s desk, or worse, in the finished part.
Pen 2.1 runs in every new AI review, no setup needed. You can try it by uploading a PDF of one of your drawings.
Sources
- Zhang, H., Liu, K., Chen, M., Li, L., Yang, S., Peng, C., & Chen, H. (2026). BenchCAD: A Comprehensive, Industry-Standard Benchmark for Programmatic CAD. arXiv preprint arXiv:2605.10865.
- Chart: benchcad.com, IoU-score by release date, retrieved 2026-09-08.
