Hugging Face's new open model chats with your DNA — on your laptop

Hugging Face released Carbon-X, a 14-billion-parameter open model that answers plain-English questions about DNA — what a mutation does to gene activity or health — and is small enough to run on a laptop, so no genomic data leaves the machine. It reads DNA in six-letter chunks, about 50,000 letters per query, and its training set deliberately excluded viruses.

Its creators caution the predictions have not been validated in wet-lab experiments. Built with Zhongguancun Academy and the University of Naples Federico II on 300 billion tokens of DNA plus English text, it is pitched as a tool for drug design and genetic therapy — the first language model that chats with the genome itself.