The path to more capable intelligence depends on the foundations beneath it. Architecture, training, memory and inference shape what systems can learn, how they reason, and who can use them.
Questions we’re interested in
- Which architectural choices support generalization, reasoning and persistent state?
- How should we measure quality alongside latency and computational cost?
- How can open research make methods easier to inspect, reproduce and improve?
Our starting point
Our direction spans model design and the systems required to train, evaluate and deploy it. Open models are part of our ambition to make reusable foundations available for others to study and build on.
What this means today
We have not published architecture results or efficiency benchmarks. Omida R1 is a separate open-weight research release, and its card does not establish those results.
Open models