Interactive

The genomic language model landscape, 2020-2026

Every notable genomic language model since 2020, by release date and parameter count, coloured by how it reads DNA. Hover any point for detail, click it for the primary source.

GenomicsLLMMachine Learning

Six years of genomic language models

Release date vs. model size. Colour shows how each model reads DNA. Hover a point for detail, click it for the source. Click any bar below to filter the chart.

Tokenization Click any colour or bar to isolate it

BHow they read DNA

k-mer and BPE gave way to reading one base at a time. Learned tokenization is the 2026 bet.

4 5 3 0 2020-23 3 0 0 1 2024 5 1 2 0 2025 2 1 0 2 2026

CArchitecture family

State-space and hybrid models got the headlines. The plain Transformer never actually left.

Transformer 19 Hybrid 4 State-space 4 Convolutional 2

29 models · Release = first public preprint or model release. Size = largest reported checkpoint; all have public weights except EDEN. GPN-MSA and GPN-Star use whole-genome alignments, so they are not directly comparable with single-sequence models. A ringed point has a cheat sheet, linked from its tooltip.