AI news · October 12, 2026
Carbon-A annotates genomes with a 1.2B-parameter open model
Carbon-A is a 1.2B-parameter genome-annotation model with a 98,304-base context window and 4.6GB weights; research teams can run it locally, but it is not a general biology chatbot.
- parameters
- 1.2B
- context window
- 98,304 bp
- reported macro nucleotide F1 on 42 genomes
- 0.944
Carbon-A is an open model for DNA annotation, not a chat model with biology branding. The Hugging Face release describes 1.2 billion parameters, a 98,304-base context window and about 4.6GB of weights. Its evaluation covers 48,167 assemblies, 22,617 taxa and 566 million predicted loci; on 42 genomes, the reported macro nucleotide F1 is 0.944. The model is designed to identify gene structure across long genomic sequences and can run from the Hugging Face stack.
The scale is useful for researchers who need a repeatable annotation pass without sending sequence data to a hosted service. It is not the same as proving protein function, disease relevance or wet-lab behaviour. The authors describe validation of some transcripts, while broader biological interpretation remains open. The comparison is with established annotation pipelines: Carbon-A may add a strong candidate pass, but every result still needs domain review and a reproducible reference set.
What you can do with it
Use the model card and start on a small, public genome sample. Measure GPU memory, runtime and agreement with your current annotation pipeline before processing a full collection. Store model version, input assembly and output coordinates so another researcher can reproduce the result.
Our take
Carbon-A is a good example of an open model winning through a narrow job and useful context length. The headline F1 is meaningful only with its stated data and evaluation design, and the model does not answer the biological questions that come after annotation. Its commercial value is in speeding a well-defined research step, not replacing a lab.
Source: Hugging Face Blog — Made With Models writes the brief; the reporting is theirs.