Hugging Face's Carbon-X reads DNA and answers in plain English

Carbon-X is an open 14.8-billion-parameter model trained on both DNA sequences and text. You paste in DNA and ask questions in a normal chat. It isn't approved for medical use.

Published October 25, 2026 9:18am EDT | Updated October 25, 2026 11:00am EDT · 1 min read

Add ScoviAI on Google

Diagram of the structure of DNA (Zephyris, CC BY-SA 3.0, Wikimedia Commons) (Hugging Face)· 1 min read

Key points

Open model from Hugging Face and research partners, built on Qwen3-14B
Trained jointly on DNA sequences and natural-language text
Can flag likely harmful mutations, find gene-control regions and explain its reasoning
Not clinically validated: not for medical decisions

Hugging Face has released Carbon-X, an open AI model that understands both DNA and English. You can paste a raw DNA sequence into a chat, ask a question about it, and get an answer in plain language.

It was built with research partners including the Zhongguancun Academy and the Telethon Institute of Genetics and Medicine in Naples.

How Carbon-X works

It has 14.8 billion parameters and starts from Alibaba's Qwen3-14B model.
DNA and text go through one shared model, with no separate DNA reader.
A special tokenizer reads text normally and reads DNA in chunks of 6 letters (6 base pairs per token).

What it can do

According to its model card, Carbon-X can:

predict whether a mutation is likely harmless or harmful,
find regions that control when genes switch on,
tell which group of organisms a sequence comes from,
predict the next letters of a sequence,
and explain its reasoning in English.

The developers report an average score of 66.7% across five DNA tests. They also list weak spots: splice sites, some protein-related questions, exact counting and positions, and generating long DNA.

An important warning

The model card is explicit: Carbon-X has not been clinically validated and should not be used for medical decisions. It's a research tool, not a diagnostic test.

Open and runnable

The model's weights are public and fit on a single 80 GB GPU. Hugging Face also publishes compressed versions for running it with llama.cpp. Because it's open, labs can test it, check it and build on it. If you're new to the idea, see Open-weight AI models, explained.

What it means for you

For most people, nothing changes today. For biology researchers and students, Carbon-X offers a free way to ask questions about DNA in plain English, as long as its answers are checked like any other unverified result.

Sources

Carbon-X model card (Hugging Face)
Carbon-X collection (Hugging Face)

Source: Hugging Face

ScoviAI Team is the ScoviAI newsroom. We write every article ourselves from the original announcements and link our sources. Spotted a mistake? Tell us and we'll fix it.