Hugging Face has introduced Carbon-X, an open-source language model designed to analyze DNA sequences. The tool allows users to input genetic data and receive explanations regarding the potential harm of specific mutations. Additionally, the model can generate new DNA sequences, which may assist in gene therapy research.
Built on the Alibaba Qwen3-14B architecture, Carbon-X features approximately 14.8 billion parameters. It was trained on 302 billion tokens, including biological materials, English text, and DNA sequences. The project was developed in collaboration with the Telethon Institute of Genetics and Medicine, the University of Naples Federico II, and the Zhongguancun Academy.
The model is released under the Apache 2.0 license, enabling researchers to run it on their own hardware. Its weights fit onto a single GPU with 80 GB of memory, ensuring that sensitive genetic data does not need to be sent to external servers. Viruses were excluded from the training data to mitigate potential misuse.
In internal benchmarks, Carbon-X achieved a score of 66.7%, trailing the OpenAI GPT-6.1 Sol model, which scored 67.5%. Hugging Face notes that Carbon-X offers a significantly lower cost per task. The company emphasizes that the model has not undergone clinical validation and is intended solely for research purposes.
The announcement follows news of Nvidia's planned acquisition of Hugging Face for $12.9 billion. The companies expect to finalize this transaction during the first half of 2027.
