Hugging Face releases Carbon-X: an open 14.8B model that chats with DNA

#What happened

Hugging Face released Carbon-X, an open 14.8-billion-parameter language model trained jointly on DNA sequences and natural language. You paste raw DNA into an ordinary chat conversation and it answers in English: predicting whether mutations are benign or pathogenic, finding regions that control gene activity, classifying sequences by organism group, and predicting the next bases of a sequence. It was built with the Zhongguancun Academy and the Telethon Institute of Genetics and Medicine / University of Naples Federico II, trained on 302 billion tokens, and ships under the Apache 2.0 license. The weights fit on a single 80GB GPU in bfloat16, so labs can keep sensitive sequences on their own hardware.

#Why it matters

Across five DNA benchmarks Carbon-X averages 66.7 percent, close to GPT-6.1 Sol at 67.5 percent, at about 1/1,800 of the inference cost per problem. Hugging Face also ships a ready-made skill so coding agents including Codex CLI can call it as a tool. Caveats straight from the model card: it has not been clinically validated and should not be used to make medical decisions, and it can miscount bases or invent a sequence's identity when context is thin.

Verified 2026-10-22. Sources: https://huggingface.co/HuggingFaceBio/Carbon-X , https://www.siliconreport.com/carbon-x-integrates-dna-and-natural-language-in-14b-model , https://madrobot.blog/2026/10/22/hugging-face-carbon-x-open-ai-model-chat-with-dna/