Hugging Face releases Carbon-X open-source DNA language model

Hugging Face has launched Carbon-X, an open-source AI model designed to allow researchers to interact with DNA sequences using natural language. Developed in collaboration with the Zhongguancun Academy in China and the Telethon Institute of Genetics and Medicine at the University of Naples Federico II, the 14.8-billion-parameter model enables scientists to paste raw DNA into a chat interface and ask questions regarding mutations, gene activity, and potential disease risks.

Built upon Alibaba’s Qwen3-14B architecture, Carbon-X utilizes a hybrid tokenizer to process DNA bases in groups of six. The model was trained on 302 billion tokens, combining genetic code with biological and English text. This approach allows the model to explain whether a genetic variant appears benign or pathogenic in plain English.

Key advantages include the ability to run locally on a single 80GB GPU, which allows laboratories to process sensitive genetic data without uploading it to external clouds. Hugging Face estimates that checking the ClinVar database for variants of uncertain significance could cost as little as $300 with Carbon-X, compared to significantly higher costs using frontier models. However, the company notes that the model has not been clinically validated, is not intended for medical decision-making, and may still struggle with specific biological markers like splice sites.