---
library_name: transformers
pipeline_tag: image-text-to-text
license: other
---

**Nanodream 3(Preview)** is an vision language model with a mixture-of-experts architecture (9B total parameters, 2B active). This model makes no compromises, delivering state-of-the-art visual reasoning while still retaining our efficient and deployment-friendly ethos.