Hugging Face gives out Carbon-X, a DNA model for a laptop

Hugging Face gave out Carbon-X on Thursday. It is an open source model that reads DNA and answers questions in English. Researchers from Zhongguancun Academy in China and the University of Naples Federico II in Italy worked on it. Lewis Tunstall of Hugging Face tells that it can help to make drugs and gene therapies.

How it works

The team started with the Qwen3-14B language model. They taught it to read DNA in groups of six letters. They trained it with 300 billion tokens (small pieces of data) of DNA and text. A user can ask if a change in DNA raises the risk of disease. The model gives a prediction and a cause.

Limits

The model reads about 50,000 base pairs, so a user cannot put in a full genome. The training data has no viruses, to limit misuse. Tunstall tells that no data goes to Hugging Face when the model operates on a local computer.

What is not known

The developers did not check the model with lab tests, so they cannot tell how accurate it is.

Insight: An open model that avoids large labs

Tunstall tells that the aim is to stop biology researchers from a need for one outside provider. In September, Nvidia agreed to buy Hugging Face for almost $13 billion. This reads as a plan to make small local models the first selection for private data, while the owner of the platform changes. Look at who owns the open tools you use.