Studio, Schedulers and 2.5 Million Evals a Week: Inside Mercor's Eval Infrastructure and Stack

Mercor runs millions of model evaluations weekly on its Studio platform, and its engineering team recently wrote about a 20x throughput win from a new compute scheduler. This module covers the eval infrastructure, the RL-environment work, the published stack from job postings, and the research on long-horizon coding - so technical candidates can speak the team's language.