Hugging Face has released Carbon-X, an open-source AI model designed to analyze DNA sequences and answer questions about them in natural language. The model, which contains 14.8 billion parameters, is based on the Qwen3-14B architecture and was developed in collaboration with the Zhongguancun Academy, the Telethon Institute of Genetics and Medicine, and the University of Naples Federico II.
The model was trained on 302 billion tokens, including DNA sequences from various organisms and scientific biological texts. It allows users to query genetic data without programming skills, identifying pathogenic mutations and regulatory elements. Carbon-X can run locally, ensuring that sensitive genetic information remains on the user's hardware rather than being transmitted to external servers.
In performance tests, Carbon-X achieved an average score of 66.7%, outperforming other open-source models. Hugging Face estimates that the model is significantly more cost-effective than proprietary alternatives; for instance, analyzing 2.4 million mutations from the ClinVar database costs approximately $300 with Carbon-X, compared to $32,000 to $4 million using other advanced models.
Despite its capabilities, the model has limitations, including a requirement for a GPU with 80 GB of memory for uncompressed operation and potential inaccuracies in counting nucleotides or determining precise coordinates. The developers emphasize that Carbon-X has not undergone clinical validation and is not intended for medical decision-making or generating new DNA sequences.
