Hugging Face releases open-source Carbon-X AI model for DNA research

What happened

Hugging Face has released Carbon-X, a new open-source AI model for understanding DNA. According to WIRED, the release took place on a Thursday, in an article dated 22 October 2026 [1]. Users can ask questions in plain language about how changes to DNA affect gene activity and health, and receive conversational answers [1]. Tunstall, a senior research scientist at Hugging Face, said the team had “unlocked a new capability” in language models, which he described as being able to chat with DNA [1].

Carbon-X was developed with researchers at Zhongguancun Academy in China and the University of Naples Federico II in Italy. It is trained on genomic sequences and natural language [1]. WIRED describes it as a kind of ChatGPT for DNA, because it combines DNA and natural language in one conversational interface [1].

The team began with an existing language model and expanded its vocabulary so it could read DNA in groups of six letters. It then continued training on 300 billion tokens, which combined DNA sequences, English text and examples linking DNA to biological descriptions [1]. The DNA sequences came from mammals, including humans, other vertebrates, invertebrates, plants, fungi and protozoa. Tunstall said the team did not train the model on viruses because it did not want the model used for potentially dangerous applications [1].

WIRED places Carbon-X among other recent work. Google DeepMind released AlphaGenome last year. It predicts the effects of DNA sequences and mutations, but in a numeric format, and it requires more compute power. Evo 2, developed by the Arc Institute and Nvidia with academic partners, understands the language of DNA and can generate entire genomes from scratch [1].

The developers of Carbon-X say they have not yet validated the model with actual wet lab experiments, so they cannot say how accurate its predictions are [1].

Hugging Face itself is described by WIRED as essentially a GitHub for AI models. It provides a library of ready-made AI tools and a platform for running mainly open-weight and open-source applications, for developers, researchers, businesses and people who want to experiment with AI [1]. In September, Nvidia agreed to acquire the company for nearly $13 billion [1]. WIRED also notes that Hugging Face was at the centre of a major hack involving OpenAI’s models earlier this year [1].

A separate Hugging Face listing describes OpenMed-NER-DNADetect-SuperClinical-184M, a model for biomedical entity recognition. It identifies proteins, DNA, RNA, cell lines and cell types in clinical texts, research papers and healthcare documents. The page says it is a fine-tuned transformer and is evaluated on the JNLPBA corpus, a 2004 benchmark built from MEDLINE abstracts with five entity types [2]. The page includes a comparison chart of the open-source model against closed-source models, but the text gives no figures for it [2]. It can be installed with pip, using the transformers and torch packages [2].

Who is affected

Tunstall said the tool is meant to help researchers design new drugs and genetic therapies. He also said it could potentially help patients interpret their own genomes [1]. One use the sources describe is helping scientists judge whether a genetic mutation is benign or disease-causing, which is a key step in developing a targeted drug or genetic therapy. The sources note that if an intervention targets the wrong mutation, it could be ineffective or could cause unintended harm [1].

Researchers face restrictions on uploading identifiable or sensitive patient data to frontier AI models. For this reason, Tunstall said the team intentionally designed a small model that runs on the user’s own hardware rather than relying on cloud-based models [1].