Mercor connects frontier AI labs with domain experts for the work that happens after pre-training. That covers reinforcement learning from human feedback, supervised fine-tuning data, model evaluation, rubric development and building reinforcement learning environments. Its expert pool spans investment banking, medicine, law and consulting, with over 30,000 specialists on the platform.
The company started in January 2023 as a recruiting marketplace. Brendan Foody, Adarsh Hiremath and Surya Midha built the first version manually over WhatsApp and Google Sheets, charging a client 500 dollars a week for a developer and keeping roughly 30 percent. The business has since moved from placing engineers to supplying the human judgement that frontier models are trained and graded against.
Vetting runs through a 20 minute structured video interview conducted by an AI voice agent, which scores technical depth, reasoning and communication. Candidates can retake it up to three times, and a passed assessment carries across to compatible roles rather than being repeated for every project. Mercor also publishes the APEX benchmark family and the open-source Archipelago evaluation harness.
Pricing is enterprise and negotiated. Mercor pays contractors directly and bills clients a marked-up hourly rate, with contractors averaging 85 dollars an hour and a reported markup of around 35 percent. There is no published list price, and permanent placements are charged separately as a share of first-year compensation.