Carbon-X is an open 14.8B single-decoder model that places DNA sequences and natural language in one token stream. Apache-2.0 weights, code and five benchmarks are public; the publisher reports a 66.7 average and single-80GB-GPU serving. It is not clinically validated and should not perform exact coordinate operations.
Carbon-X opens a DNA-and-language model for local research
2026-10-21Verified announcementTRIAL
Carbon-X is an open 14.8B single-decoder model that places DNA sequences and natural language in one token stream. Apache-2.0 weights, code and five benchmarks are public; the publisher reports a 66.7 average and single-80GB-GPU serving. It is not clinically validated and should not perform exact coordinate operations.
Availability: Public weights; vLLM or Transformers; one 80GB GPU in bfloat16
What this means for you
For bioinformatics researchers, local weights can keep sensitive sequences off third-party services and lower experimentation cost. Publisher comparisons against frontier models are not independently reproduced and do not support clinical use.
Evidence & related material
Technical details & scores
| Parameters | 14.8B |
|---|---|
| Architecture | Single decoder with a hybrid text and DNA tokenizer |
| License | Apache-2.0 |
| Training | 302B mid-training tokens |
| Context Window | 40,960 tokens |
| Availability | Public weights; vLLM or Transformers; one 80GB GPU in bfloat16 |
| Publisher Evaluation | 66.7 average across five DNA benchmarks; one sample per problem; comparison reasoning levels vary |
| Limitation | Not clinically validated; exact base counting and coordinates require deterministic tools |
| Time Basis | Hugging Face repository created 2026-10-21T10:21:04Z; recovered by recall outside the overlap. |
Technical impact 4
Engineering value 4
Adoption signal 2
Maturity 2
Verification cost 3
Risk 4