Research

Research and evaluations.

We build models for agent workloads and evaluate them on the work agents actually do — tool use, verification, and research — not leaderboard trivia.

Evaluations in progress.