Eunoia Omega
Abstract and interactive reasoning at sub-billion scale, with a pre-registered claim about cost falling as experience accumulates.
The core lab. Every model on this site was designed here, and most of them were broken here first, on purpose, before anyone else saw them.
Tech & Research owns the intelligence itself — the models, the research programmes behind them, and the standards that decide when something is ready to leave the lab.
It is deliberately slow. Work here is measured against whether a claim survives its own falsification test, not against a release schedule.
Abstract and interactive reasoning at sub-billion scale, with a pre-registered claim about cost falling as experience accumulates.

A system built around an internal state that evolves on its own, where the settling of that state is the thinking.

A world model that predicts what happens next in a situation rather than what word comes next in a sentence.
Nothing moves to AI Systems on enthusiasm. A programme ships when it has results that survive the evaluation protocol written before it started, and it is abandoned when it hits a threshold that was also fixed in advance.
Both outcomes are published. A claim that fails its own test is still a finding, and burying it would make every other number we report worth less.