Hi! I wanted to share my new blog on the costs of running AI Evals. We dig into how benchmarking frontier systems now routinely costs tens of thousands of dollars per run, why agent evals are especially unpredictable, and what that concentration of validation authority means for the broader research community.