Hugging Face releases Carbon-X, an Apache 2.0 open DNA model that runs on a laptop
Article timeline
The articles in this story. Times are ET.
Lewis TunstallXLewis Tunstall asks whether to scale Carbon-X's 14B recipe to 100B+ parametersAILewis Tunstall of Hugging Face asks whether the recipe behind the 14B-parameter Carbon-X model should be scaled to 100B+ parameters. The post builds on a Hugging Face Bio announcement describing Carbon-X as fluent in English and DNA, and the model is available on Hugging Face.
Leandro von WerraXCarbon-X model claims frontier-level DNA understanding and English-DNA chatAIHugging Face's Leandro von Werra says Carbon-X is the first model to reach frontier intelligence on DNA understanding. He says it handles both English and DNA, so users can chat with DNA sequences. The model is published on Hugging Face under HuggingFaceBio/Carbon-X.
- MarkTechPostNewsHugging Face releases Carbon-X, an open 14.8B DNA language model that reads DNA and answers in English
AIHugging Face has released Carbon-X, an open-weight 14.8B-parameter model that reads raw DNA and answers in English, under Apache 2.0. It averages 66.7% on five DNA benchmarks, self-reported, and runs on one 80 GB GPU, while scoring 29.8% on GenomeQA splice-site questions.
- Lewis TunstallXCarbon-X and llama.cpp let users chat with their DNA locally
AILewis Tunstall says Carbon-X and llama.cpp let users chat with their DNA on their own machine. The post gives no further technical details, pricing, or model specifics.
- Lewis TunstallOfficialHugging Face releases Carbon-X, an open 14B model that reads raw DNA
AIHugging Face says it released Carbon-X, an open 14B model that reads raw DNA and explains it in plain English, built with TIGEM, the University of Naples Federico II, and Zhongguancun Academy. It reports 67% on its DNA benchmarks, versus 61% for Claude Opus 5.5 and 50% for Gemini 4 Argon, and says it runs on a laptop. The post says human and animal viruses were excluded from training data and gives an NF1 splice-site example where the model calls a mutation pathogenic.
Thomas WolfOfficialPickHugging Face releases Carbon-X, an open 14B model that reads raw DNAAIHugging Face says it released Carbon-X, an open 14B model that reads raw DNA and explains it in plain English, built with TIGEM, Naples Federico II University and Zhongguancun Academy. It reports a 67% score on its DNA benchmarks versus 61% for Claude Opus 5.5 and 50% for Gemini 4 Argon, within one point of GPT-6.1 Sol at about 1000x lower cost. Human and animal viruses were excluded from training data, and the model runs on a laptop.
- Leandro von WerraOfficialHugging Face releases Carbon-X, a Qwen3.5 14B model trained to reason over DNA
AILeandro von Werra says the Hugging Face team mid- and post-trained Qwen3.5 14B on natural language and DNA sequences to reason over DNA. He claims the model beats other open models and performs on par with the best frontier models, at a fraction of the cost and without distillation. The model is available as Carbon-X on Hugging Face, and the full recipe is being released.
- Lewis TunstallOfficialLewis Tunstall discusses Carbon-X, a local AI model for DNA
AILewis Tunstall of Hugging Face thanked WIRED's Emily Mullin for a conversation about the release of Carbon-X. WIRED describes Carbon-X as a biology AI model that can be run locally on a laptop.
clem 🤗OfficialPickHugging Face releases Carbon-X, an open-source DNA model built on Qwen3-14BAIHugging Face's Clément Delangue says Carbon-X, a DNA model, was made by midtraining Qwen3-14B on 300B tokens of DNA, English and interleaved data. He says it outperforms all open models evaluated on five DNA benchmarks, a 27-point gain over Qwen3-14B, and reaches API frontier level at 1,800x less cost than GPT 6.1 Sol. The model is released under Apache 2.0 on Hugging Face.
- Lewis TunstallOfficialHugging Face releases Carbon-X, an open model trained on DNA and text
AILewis Tunstall announces Carbon-X, an open DNA model that Hugging Face says was midtrained from a 14B model on DNA and text. He says it beats every open language model and matches GPT-6.1 Sol at 1,000x lower cost. Users can ask it about sequences and mutations or have it generate new DNA. WIRED describes the model as a ChatGPT-style conversational interface for DNA.
- Lewis TunstallOfficialPickHugging Face releases Carbon-X, a 14B model that combines DNA and natural language
AIHugging Face releases Carbon-X, a 14B model trained on 300B tokens of interleaved DNA and text, followed by SFT and RL. The author says it beats every open language model tested on DNA benchmarks and comes within a point of GPT-6.1 Sol at over 1,000x lower inference cost. The model is Apache 2.0 licensed and runs locally on a laptop with llama.cpp.
- Wired · AINewsHugging Face releases Carbon-X, an open-source model for chatting with DNA
AIHugging Face released Carbon-X, an open-source model that answers natural-language questions about DNA, developed with researchers at Zhongguancun Academy and the University of Naples Federico II. The 14-billion-parameter model can run locally on a laptop, so no data goes to Hugging Face or another company. The developers say they have not yet validated its predictions in wet lab experiments.