Mercor raises $750 million at a $20 billion value to test AI agents

Mercor supplies the human experts who train and test AI models. It has more than 5 million experts. It raised $750 million at a $20 billion value in a Series D round, a funding round. In October 2025, a round of $350 million valued it at $10 billion. Mercor says annual revenue passed $2 billion in June.

What the money buys

Mercor says the money pays for research and test systems, with more practice environments for agents. Mercor bought the startup Deeptune in July. Deeptune recreated hundreds of company applications for agents to practice on. For each customer, Mercor makes a custom benchmark, a test based on the rules of that company.

Who invests

Nvidia, Salesforce and Zoom put in money. Benchmark, Felicis and General Catalyst also took part. Mercor named no lead investor.

What is not known

The round includes a tender offer, which lets employees sell shares. Mercor has not said how the $750 million divides between that and new money. Adarsh Hiremath, co-chief executive, says a model can score high on a public benchmark. It can still be wrong on the policies of one business.

Insight: Companies will pay to test agents on their own work

Mercor sold mostly to AI labs. It paid for Deeptune, which makes practice copies of company applications. Earlier reports say that iMerit also sells expert tests. This looks like a race to own the test that shows if an agent can do a task, and the owner of the test sets the standard. Select the test that you will use.