Results from the ELO Lab.
We build artificial minds that run on small hardware and keep learning. This is where we publish what they can do.
Our rule: we publish what we built. Not how.
Published
Model
The Golden model
A byte-level language model that rewrites its own numbers as it learns, and keeps learning after it ships.
Measurement
The whole test
One text-only exam across five areas of intelligence — verbal, reasoning, quantitative, memory, spatial — scored against adult norms.
Architecture
JEPA + LLM, as one
A model trained to predict the next letter and the meaning of what comes next, in the same step. Two ideas, one architecture.