Hugging Face Bio releases Carbon-X, a 14.8B open DNA model
First merged summary.
Hugging Face Bio’s Carbon-X, an open DNA-and-English model, is public with publisher-reported benchmark scores.
Hugging Face Bio released Carbon-X, a decoder-only model started from Qwen3-14B and trained jointly on DNA and natural-language text. It answers DNA questions in English, including variant-effect and next-base tasks, using a hybrid tokenizer that encodes DNA as 6-mers. The publisher reports 66.7% across five DNA benchmarks with thinking disabled, 27 points above Qwen3-14B and just under a cited GPT-6.1 Sol score, at far lower inference cost. The model card lists 14.8 billion parameters; MarkTechPost describes Carbon-X as a 14-billion-parameter model. Weights are released for local serving with Transformers v5 or later and vLLM.
- Architecture is 40 layers, hidden size 5120, and a 40,960-token context.
- Mid-training used 302B tokens, then supervised fine-tuning and reinforcement learning.
- Publisher reports 66.7% versus 67.5% for GPT-6.1 Sol at about 1/1,800 the cost.
- Lab estimates analysing ClinVar’s 2.4 million uncertain variants would cost about $300.
- It fits on one 80 GB GPU in bfloat16, according to the publisher.
Timeline
dated events in the reporting · oldest first- Oct 21, 2026Carbon-X listed trending at number 29 · Hugging Face trending models
- Oct 25, 2026MarkTechPost reports the Carbon-X release · MarkTechPost
What each outlet adds
beyond the shared facts- Hugging Face trending modelsArchitecture, training scale, benchmarks, cost estimate, and serving requirements.
- MarkTechPostCalls it a 14B release; captured text lacks training and license details.
- Organizations
- Hugging Face BioHugging Face
Coverage
oldest first · one row per article · UTCHugging Face Bio releases Carbon-X, a 14.8B DNA model
The open chat model, started from Qwen3-14B, reports 66.7% on five DNA benchmarks with thinking disabled.
Hugging Face releases Carbon-X, a 14B open DNA language model
The announcement frames Carbon-X as an open model that jointly handles DNA sequences and natural language.