The full task set

Public and private halves, for teams that evaluate internally. We ask who you are so the private half stays private.

A model on the board

We run it under the same protocol as every other model and report the cost back to you.

Research cooperation

Shared task families, held-out evaluation of your training runs, joint write-ups.

A benchmark for you

Private benchmarks from your own codebases and workflows, for choosing models, routers or agents.