Arc
A low-cost, experimental English co-writer. 120.8M parameters, a 1,024-token context, and published writing samples alongside its known limitations.
Start with Arc.
A compact model with a published evaluation. Meridian and Zenith remain roadmap directions, not evaluated releases.
Meridian
The next model direction: develop language and code capability while keeping local hardware central. No evaluated checkpoint or release specification is published here yet.
Zenith
A direction for higher-capacity Geocentric models. Architecture, release timing, and capability remain research questions, not product commitments.
CURRENT SMALL-MODEL WORK
120.8M parameters. 1,024-token context.
11transformer layers
896model width
32Kvocabulary
1 GPUtraining target
Small models are the proving ground.
Geocentric is using compact models to test architecture, training efficiency, data quality, post-training, evaluation, and inference before scaling the same ideas further.
The goal is simple: make each generation meaningfully more capable without losing the hardware efficiency that defines the project.
See the systems behind the models