Key facts
- Carbon-X allows scientists to ask questions about DNA sequences and receive predictions and explanations.
- The model can also generate new DNA sequences for gene therapy development.
- Carbon-X was trained on genomic sequences from humans, mammals, plants, and fungi.
- Viruses were excluded from the training data to prevent misuse.
Hugging Face has launched Carbon-X, a new AI model designed to enable scientists to analyze and generate DNA sequences through natural language interactions. The model, announced on Thursday, aims to democratize access to advanced AI tools for biological research, preventing over-reliance on major AI laboratories. Lewis Tunstall, a senior research scientist at Hugging Face, explained that researchers can use Carbon-X to query DNA sequences, such as assessing disease risk associated with specific mutations, and to generate novel sequences for gene therapy development. The model was built by training an existing language model, Qwen3-14B, on a diverse dataset of genomic sequences from humans, other mammals, plants, and fungi, with viruses explicitly excluded from the training data to mitigate misuse. Hugging Face, which is being acquired by Nvidia, developed Carbon-X in collaboration with research labs in Italy and China, making it an open-source tool that can be run on local computers to protect sensitive patient data. Tunstall suggested that such specialized models are less prone to the misalignment issues observed in larger, generalist AI systems.