Meet Krus, our fast everyday model
The smallest model in the Ichnus family does the most work. Here is why a fast tier matters for free access.
Most conversations with Ichnus are answered by Krus, the smallest and fastest model in the family — and most people never notice. That is the point.
Krus is not an afterthought to the “real” models. It is the workhorse that makes free access sustainable. Every request Krus can answer well is a request we do not have to answer with a much more expensive model, and that arithmetic is what lets a non-profit serve a free assistant to everyone rather than a paid one to some.
What Krus is for
Krus handles the everyday: quick factual questions, short drafts, simple edits, routine lookups. It is the lowest-latency model in the family, tuned to be fast and cheap to serve. For the user, it feels like an assistant that thinks instantly.
It is also deliberately modest. Krus will decline tasks it cannot do well — or hand them up to Mantus. The routing system we built (documented in our tiered routing paper) means you get the right model for the question without thinking about it.
What Krus is not for
We do not pretend Krus is a frontier model. It is not the tool for deep research, complex multi-step reasoning, or long-document work — those are Mantus and Corus jobs. What matters is that the boundary is honest: Krus knows what it is good at, and the system knows when to route around it.
Why the small model matters
There is a version of the future where powerful AI is free because the powerful model is cheap. We are working on that (see our efficiency work). There is another version where free access exists because most questions are answered by a small, efficient model — and the big one is saved for the questions that need it.
Both versions are true at Curos. Krus is the second one, and it is the quiet reason Ichnus can be free for everyone.
All news →