<!-- mobian-agent-page publisher="dailydev" canonical="https://daily.dev/posts/jFrgS8pIQ" -->

---
title: Hugging Face releases Carbon-X, an open-source DNA...
description: Hugging Face released Carbon-X, an open-source model that answers plain-language questions about DNA sequences. It runs on a single 80GB GPU under the Apache...
canonical: https://daily.dev/posts/hugging-face-releases-carbon-x-an-open-source-dna-language-model-jfrgs8piq
twitter:card: summary_large_image
twitter:site: @dailydotdev
og:type: website
og:site_name: daily.dev
og:title: Hugging Face releases Carbon-X, an open-source DNA language model | daily.dev
og:description: Hugging Face released Carbon-X, an open-source model that answers plain-language questions about DNA sequences. It runs on a single 80GB GPU under the Apache...
og:url: https://daily.dev/posts/hugging-face-releases-carbon-x-an-open-source-dna-language-model-jfrgs8piq
og:image: https://api.daily.dev/og/posts/jFrgS8pIQ.png
og:image:alt: Hugging Face releases Carbon-X, an open-source DNA language model
og:image:width: 1200
og:image:height: 630
og:locale: en
---

> ## Documentation Index
> Fetch the complete documentation index at: https://daily.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Hugging Face releases Carbon-X, an open-source DNA language model

**[Collections](https://daily.dev/sources/collections)** · 3 min read · 1 upvotes · 0 comments

## Summary

Hugging Face released Carbon-X, an open-source model that answers plain-language questions about DNA sequences. It runs on a single 80GB GPU under the Apache 2.0 license. Built from Alibaba's Qwen3-14B with about 14.8 billion parameters, it was midtrained on roughly 302 billion tokens of DNA, English text and biology data, using a hybrid tokenizer that splits English into BPE and DNA into 6-mers. It beats every open model Hugging Face tested on five DNA benchmarks, a 27-point gain over its Qwen3-14B base, and scores 66.7% on Hugging Face's own tests, just behind OpenAI's GPT-6.1 Sol at 67.5%. The company says it costs about 1,800 times less. The model card says it has not been clinically validated and should not be used for medical decisions.

## Content

Hugging Face has released Carbon-X, an open-source language model that reads raw DNA sequences and answers questions about them in plain English. It's built on Alibaba's Qwen3-14B and has about 14.8 billion parameters. Weights are under the Apache 2.0 license.

## What it is

Hugging Face midtrained Qwen3-14B on roughly 300 billion tokens of DNA, English text and interleaved data. It uses a hybrid tokenizer: BPE for English, a 6-mer tokenizer for DNA. The model can predict whether a mutation is pathogenic, flag regulatory regions, classify organisms and predict next bases.

The training work was done with TIGEM (Telethon Institute of Genetics and Medicine), Naples "Federico II" University and Zhongguancun Academy. Human and animal viruses were kept out of the training data.

It runs on a single 80GB GPU, or as a 15.8GB Q8_0 GGUF. @Thom_Wolf describes the hardware as within reach of academia, down to a laptop.

## Benchmarks

Across five DNA understanding and generation benchmarks, Carbon-X averages 66.7%. That's a 27-point gain over the base Qwen3-14B and ahead of every open language model Hugging Face evaluated. These are the company's own tests.

- GPT-6.1 Sol: 67.5%, so Carbon-X is 0.8 points behind
- Claude Opus 5.5: 61%
- Gemini 4 Argon: 50%
- BioReason Coding: Carbon-X leads at 74.9%

Splice-site accuracy is weak, at 29.8%.

Cost is the bigger gap. Hugging Face says scoring 2.4 million ClinVar variants would cost about $300 with Carbon-X, versus $32,000 to $4 million on frontier models. @ClementDelangue puts the difference against GPT-6.1 Sol at 1,800x; Wolf says about 1,000x.

One example from the launch blog: given an 8kb stretch of the NF1 gene with a mutation at a splice site, most frontier models call it benign, refuse, or run out of tokens. Carbon-X calls it pathogenic, because the mutation breaks the splice site.

## The argument behind it

The release is partly a complaint. Delangue says frontier labs are advancing AI for biology while blocking scientists, academics, non-profits and health agencies with broad but shallow guardrails that bad actors bypass easily. He says the filter blocked him from proofreading his own launch tweet. Wolf says his AI assistant was cut off by a bio safety filter while summarizing his team's notes on the model.

Delangue's worry is an asymmetry like the one in cybersecurity, made sharper now that frontier labs are building wet labs. His answer is to release openly so anyone can study the model's capabilities, limitations and risks.

## Limits

The model card says Carbon-X hasn't been clinically validated and shouldn't be used for medical decisions.

The release comes weeks after Nvidia agreed to buy Hugging Face for about $12.9 billion.

## Questions this post answers

### What is Hugging Face's Carbon-X DNA language model built on?

Carbon-X is built from Alibaba's Qwen3-14B base model, with about 14.8 billion parameters. Hugging Face midtrained it on roughly 302 billion tokens of DNA, English text, biology data and interleaved data, using a hybrid tokenizer where English is processed with BPE and DNA sequences with a 6-mer tokenizer. It's released under the Apache 2.0 license and runs on a single 80GB GPU.

_Developers tracking open-weight model releases can follow benchmarks like these on daily.dev._

### How does Carbon-X compare to GPT-6.1 Sol on DNA benchmarks?

Carbon-X scored 66.7% on Hugging Face's own DNA benchmark tests, just behind OpenAI's GPT-6.1 Sol at 67.5%. Hugging Face claims Carbon-X costs about 1,800 times less to run. These figures come from Hugging Face's own evaluations rather than independent testing, and Carbon-X also beat every open language model tested, including a 27-point improvement over its Qwen3-14B base.

_Anyone weighing open versus closed models for a project can track these comparisons on daily.dev._

### Is Carbon-X safe to use for medical or clinical decisions?

No, the model card explicitly states that Carbon-X has not been clinically validated and should not be used for medical decisions. It is intended as a research tool for answering plain-language questions about DNA sequences, not as a diagnostic or treatment aid.

_Teams evaluating AI models for regulated use cases can keep tabs on such disclaimers via daily.dev._

## Similar posts on daily.dev

- [Hugging Face review: Powering the future of open source AI](https://daily.dev/posts/hugging-face-review-powering-the-future-of-open-source-ai-ctuakivqo) · All Things Open · 1 upvotes · 0 comments
- [Hugging Face Releases FineTranslations, a Trillion-Token Multilingual Parallel Text Dataset](https://daily.dev/posts/hugging-face-releases-finetranslations-a-trillion-token-multilingual-parallel-text-dataset-ilbhxxjto) · InfoQ · 0 upvotes · 0 comments
- [We Got Claude to Fine-Tune an Open Source LLM](https://daily.dev/posts/we-got-claude-to-fine-tune-an-open-source-llm-syttcqt2y) · Hugging Face · 4 upvotes · 0 comments

---

Tags: [#open-source](https://daily.dev/tags/open-source), [#llm](https://daily.dev/tags/llm), [#biotech](https://daily.dev/tags/biotech)

[View this post on daily.dev](https://daily.dev/posts/hugging-face-releases-carbon-x-an-open-source-dna-language-model-jfrgs8piq)

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]}
{"@context":"https://schema.org","@type":"TechArticle","headline":"Hugging Face releases Carbon-X, an open-source DNA language model","url":"https://daily.dev/posts/hugging-face-releases-carbon-x-an-open-source-dna-language-model-jfrgs8piq","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/hugging-face-releases-carbon-x-an-open-source-dna-language-model-jfrgs8piq"},"datePublished":"2026-10-22T13:41:49.716Z","dateModified":"2026-10-27T16:38:06.102Z","description":"Hugging Face released Carbon-X, an open-source model that answers plain-language questions about DNA sequences. It runs on a single 80GB GPU under the Apache...","image":"https://pbs.twimg.com/media/HVPOfkEWAAAiUCS.jpg","thumbnailUrl":"https://pbs.twimg.com/media/HVPOfkEWAAAiUCS.jpg","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/hugging-face-releases-carbon-x-an-open-source-dna-language-model-jfrgs8piq","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"open-source,llm,biotech","timeRequired":"PT3M"}
{"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"Hugging Face releases Carbon-X, an open-source DNA language model"}]}
{"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/hugging-face-releases-carbon-x-an-open-source-dna-language-model-jfrgs8piq#faq","mainEntity":[{"@type":"Question","name":"What is Hugging Face's Carbon-X DNA language model built on?","acceptedAnswer":{"@type":"Answer","text":"Carbon-X is built from Alibaba's Qwen3-14B base model, with about 14.8 billion parameters. Hugging Face midtrained it on roughly 302 billion tokens of DNA, English text, biology data and interleaved data, using a hybrid tokenizer where English is processed with BPE and DNA sequences with a 6-mer tokenizer. It's released under the Apache 2.0 license and runs on a single 80GB GPU. Developers tracking open-weight model releases can follow benchmarks like these on daily.dev."}},{"@type":"Question","name":"How does Carbon-X compare to GPT-6.1 Sol on DNA benchmarks?","acceptedAnswer":{"@type":"Answer","text":"Carbon-X scored 66.7% on Hugging Face's own DNA benchmark tests, just behind OpenAI's GPT-6.1 Sol at 67.5%. Hugging Face claims Carbon-X costs about 1,800 times less to run. These figures come from Hugging Face's own evaluations rather than independent testing, and Carbon-X also beat every open language model tested, including a 27-point improvement over its Qwen3-14B base. Anyone weighing open versus closed models for a project can track these comparisons on daily.dev."}},{"@type":"Question","name":"Is Carbon-X safe to use for medical or clinical decisions?","acceptedAnswer":{"@type":"Answer","text":"No, the model card explicitly states that Carbon-X has not been clinically validated and should not be used for medical decisions. It is intended as a research tool for answering plain-language questions about DNA sequences, not as a diagnostic or treatment aid. Teams evaluating AI models for regulated use cases can keep tabs on such disclaimers via daily.dev."}}]}
```

