According to CoinMeta, Hugging Face has collaborated with Beijing Zhongguancun University and Italian research institutions to open-source a biological model called Carbon-X. Researchers can analyze data using this model in the same way they would use ChatGPT. In the five official benchmark tests published for this model, it achieved an average score of 66.7%, which is close to the 67.5% scored by GPT-6.1 and SOL. The foundation of this model is Qwen3-14B, which was released in 2025. The team added a vocabulary specifically designed to recognize DNA to it and continued training it with approximately 302 billion token units of DNA data, as well as biological texts and instructional information. The model has around 14.8 billion parameters in total, and its weights as well as a localized, quantized version are now available. Researchers can process sensitive genetic data on their own devices.