METR

METR (Model Evaluation and Threat Research) is an independent nonprofit AI safety organization that evaluates frontier AI models for dangerous autonomous capabilities. They conduct pre-deployment evaluations for major AI developers including Anthropic, OpenAI, and Google DeepMind under Responsible Scaling Policy frameworks. METR also publishes research on benchmark validity and real-world AI performance, including notable studies on SWE-bench merge rates and developer self-perception of AI productivity gains.