We train models end-to-end with risk control for the broad family of optimized certainty equivalent (OCE) risks.