# cadrille: Multi-modal CAD Reconstruction with Reinforcement Learning

**Date:** 2026-01-25

M:3L researchers co-authored cadrille, accepted as an Oral at [ICLR 2026](https://iclr.cc/virtual/2026/events/oral), a [CORE 2023 A*-ranked conference](https://portal.core.edu.au/conf-ranks/?by=rank&page=1&search=A%2A&sort=atitle&source=CORE2023). The model turns point clouds, multi-view images, or text descriptions into editable Python-based CAD programs within a single vision-language architecture. It combines supervised fine-tuning on large-scale procedurally generated designs with online reinforcement learning driven by programmatic feedback, improving geometric accuracy and code validity and achieving leading results across DeepCAD, Fusion360, CC3D, and Omni-CAD benchmarks. Its earlier arXiv title included ‘Online Reinforcement Learning.’

## Related links

- [Read the paper](https://arxiv.org/abs/2505.22914)
- [Official ICLR oral listing](https://iclr.cc/virtual/2026/events/oral)
- [Final published paper](https://openreview.net/pdf?id=w2tnhhMbXv)
- [View code](https://github.com/col14m/cadrille)
- [CORE 2023 ranking](https://portal.core.edu.au/conf-ranks/?by=rank&page=1&search=A%2A&sort=atitle&source=CORE2023)

## Authoritative URLs

- [Canonical news page](https://m3l.am/news/cadrille-iclr-2026)
- [All M:3L news and events](https://m3l.am/news.md)