Key Info

Hugging Face announced Carbon-X, described as the best open-source DNA model, created by midtraining Qwen3-14B on 300B tokens of DNA, English, and interleaved data. The team says it beats every open language model and matches GPT-6.1 Sol at 1,000x lower cost.

Highlights

  • Built on Qwen3-14B with midtraining on 300B tokens of DNA, English, and interleaved data
  • Positioned as the best open-source DNA model
  • Claims to outperform all open language models and match GPT-6.1 Sol at 1,000x lower cost
  • Supports asking about a sequence, predicting mutation effects, and generating new DNA
  • Framed as a response to restricted access to frontier biology AI for scientists and non-profits