Open-sourced Conversational Large Model That Can Directly Read Raw DNA: Based on Qwen3-14B

On October 25, 2026, Hugging Face released the open-source DNA language model Carbon-X. The model has a total of 14.8 billion parameters (approximately 14.8 billion), is based on Qwen3-14B for training, and uses the Apache 2.0 open-source license.

The official states that this is the first time raw DNA reading and natural language interaction have been integrated into one model. Users can directly input raw DNA base sequences along with English questions, and the model will return interpretation results in natural language format, following the chat interaction logic of typical large language models.

Previously, gene analysis AI tools generally had a high threshold. Raw sequencing data needed to undergo multiple preprocessing steps such as noise reduction, feature annotation, and format conversion before being input into the model. The output results were mostly structured professional data, which required a background in bioinformatics to understand. Cross-disciplinary developers who wanted to call related capabilities had to thoroughly study a complete set of dedicated toolchain documents.

Official test data shows that Carbon-X achieved an average accuracy of 66.7% across 5 mainstream DNA analysis benchmark tests, covering common gene analysis tasks such as sequence classification, function prediction, and variant identification.

Apache 2.0 is a highly business-friendly open-source license. Individual researchers and enterprise teams can freely use, modify, and redistribute the model, without needing to disclose business codes developed based on the model. Currently, the model weights and related documents have been launched on the Hugging Face platform, and developers with deployment conditions can directly pull and use them.

Source: MarkTechPost, Author: Asif Razzaq