Hugging Face’s new AI lets scientists chat with DNA

Hugging Face has released Carbon-X, an open AI model that answers questions about DNA in plain English.

Researchers can give it a genetic sequence and ask whether a mutation could be harmful or which regions help control gene activity. The model provides written explanations and supports follow-up questions in the same conversation.

Carbon-X builds on Alibaba’s Qwen3-14B and was trained on genetic sequences, biological information and ordinary text. Its developers expanded the model’s vocabulary to read DNA in six-letter groups, allowing it to process genetic code alongside everyday language.

Lewis Tunstall, a senior research scientist at Hugging Face, told WIRED that the tool is intended to help researchers develop drugs and genetic therapies. It could help scientists investigate mutations linked to disease and identify changes worth studying further.

The downloadable model can run on researchers’ own hardware, keeping sensitive or unpublished genetic data away from external AI services. Its standard version requires a GPU with 80GB of memory, and the weights are available under the Apache 2.0 license.

In the team’s tests across five DNA benchmarks, Carbon-X performed close to GPT-6.1 Sol and ahead of the other open models evaluated. Its predictions have not yet been validated in laboratory experiments, however, and Hugging Face warns that it can produce incorrect explanations and should not be used for medical decisions.