The data behind
the next breakthrough

A reward model trained on organic preferences from higher-quality raters across diverse tasks.

200+ evaluations and long-horizon RL environments to test your model.