The Deriva Series.

Our first public model series, built for efficiency, productivity, and computer use. In development, with no public model access yet.

Public launch targetTiming may change. Development and improvements will continue after launch.

Deriva Series card

Models in this series.

In development

Deriva-1

Our lead priority. Deriva-0.7 shows internal long-context gains, with reinforcement learning being introduced. Qwen3.8-27B is the current candidate base, not a final commitment.

What the series is for

Where Arcus narrows in on software, Deriva is built to be broad. It is the series we expect most people to use most of the time, the dependable everyday model in the lineup. Writing, analysis, vision, questions that do not fit neatly into any one specialty.

That breadth comes with a deliberate limit. Deriva is meant to be competent at code, not exceptional. Arcus is our concept for a future software specialist; it has not been built. Deriva is the general-purpose model we are developing first.

Deriva is currently our highest priority series. After reviewing how much we had running at once, we set a clear order across projects, and Deriva leads our model work: see Evaluating scope for why. Deriva-1 is also the model Kova, our AI assistant, is built around.

Related