METR

METR is a research organization focused on Model Evaluation & Threat Research that studies AI developer productivity and provides benchmark data on AI agent capabilities. The organization has conducted research identifying reward-hacking behavior in AI models.