Hugging Face Unveils Carbon-X, an AI Model for Reading DNA

Hugging Face announced the Carbon-X model on Thursday. Carbon-X is the latest and most powerful model in Hugging Face's Carbon family of AI models, which are built to analyze DNA.

With Carbon-X, users can paste in a DNA sequence and ask a question in plain English. The model then gives an interpretation, prediction or explanation. Hugging Face research scientist Lewis Tunstall told CNET that Carbon-X lets researchers "chat with DNA" for the first time, according to CNET.

Many cloud-based chatbots, such as ChatGPT, Gemini and Claude, process requests remotely (on servers somewhere else) rather than on the user's computer. Carbon-X can run on a user's own hardware, so it does not need someone else's computer to work.

Carbon-X cannot physically sequence DNA, which means it cannot read DNA from a sample. Tunstall said the model's strongest area is predicting mutation effects (how a change in DNA may affect it) and completing DNA sequences.

Carbon-X can also work with other AI agents and give another AI model access to a DNA specialist. Hugging Face says the model has limits, though. It can miscount in sequencing operations and can hallucinate (produce information that is not true). It may also invent the identity or function of a sequence when the prompt does not have enough information.

The quality of Carbon-X's answers tends to drop when users ask about particularly long DNA sequences. Carbon-X is an open model, and its training recipe (the method used to build it) is publicly released.

In September, Nvidia announced it would buy Hugging Face for a little under $13 billion. Nvidia CEO Jensen Huang called Hugging Face a "vibrant home for the open model developer community."