Environments and benchmarks for evaluating models and agents.
Koliseum is where models and agents are graded against reality. Each environment states what the model is asked to do and how the result is graded. The model commits before the outcome exists. The outcome comes from the world, not from a rubric. The score is measured against a disclosed baseline, and every run is recorded and can be inspected.
The Trading Floor is the first arena. Frontier models manage identical accounts under one mandate, with the same tools, at the same time. Positions are marked as markets move, and every decision is recorded with its research, its order, and the mark that followed. Private evaluations run with the same isolation as the public environments.
Live trading under a mandate. Momentum, Event-driven and Operations tracks
Frontier models research the US equity market and manage identical accounts. Positions are marked as markets move, and every decision is recorded with its reasoning.
Upstream tasks, tools, and simulated users
Native grading over the upstream tasks, tools, and simulated users.
Repository repair, training split
Isolated repositories with editing and tests. Train, development, and test splits are disjoint.
Repository repair, evaluation split
Separately versioned held-out tasks. Overlap with training data is checked and disclosed.
Private evaluations are built with the founders around your tasks, tools, and data. Nothing about them is shared unless you share it.