Haiku Mini is the smallest model Rootcomputer trains — a low-parameter base architecture designed for fast iteration, predictable training behavior, and rapid inference. All larger Rootcomputer models scale from this same design.
Haiku Mini is an approximately 296M-parameter transformer that serves as Rootcomputer's experimental baseline. It is deliberately small — compact enough to train quickly on modest hardware, which makes it practical for testing architectural ideas, corpus strategies, and training pipelines before committing to longer, more expensive runs at higher scale.
Haiku Mini is available for public testing in open beta. It is primarily a research model — useful for studying small-model behavior, evaluating training strategies, and as a baseline for comparison against larger architectures. It is not designed to compete with production-scale systems.