Rootcomputer

Rootcomputer is an independent AI research lab that builds and studies language models typically under 7 billion parameters. The organization designs model architectures, assembles pretraining corpora, and runs multi-phase training pipelines to study downstream behavior, alignment, and reliability at a small scale. The lab's research focuses on transformer architecture design, pretraining corpus curation, and behavioral stability in compact models. Its developed model lines include Epos, Haiku, Tanka, and Prizm, which are made available for research and open beta testing.