Hugging Face releases Carbon-X, a 14.8B open DNA language model

Hugging Face Bio has released Carbon-X, an open decoder-only DNA model built on Qwen3-14B that reads raw DNA in a chat and answers in English. Weights ship under Apache 2.0 with a 40,960-token context and run on a single 80 GB GPU; it averages 66.7% across five DNA benchmarks, the best of any open model tested.

• 14.8B parameters, 40,960-token context, runs on one 80 GB GPU
• Adds 5,515 DNA tokens: 4,096 six-mers, 1,364 tail tokens, 3 markers
• 66.7% average on 5 DNA benchmarks, 74.9% on BioReason Coding
• Weakest result: 29.8% on GenomeQA splice-site questions