Mercor raises $750 million Series D at a $20 billion valuation, selling expert networks and eval infrastructure to enterprises

Mercor announced a $750 million Series D at a $20 billion valuation, and the funding accelerates its investment in infrastructure and research for evals and post-training and in building more environments where agents can learn complex workflows. For the previous two years Mercor focused on building infrastructure to train and evaluate frontier models on real-world work for frontier labs; it is now bringing those same capabilities to enterprises across the economy. The round included an employee tender and participation from Benchmark, Felicis, Fundomo, funds managed by affiliates of Apollo, G Squared, General Catalyst, Glade Brook Capital, NVIDIA, Salesforce, Sapphire Ventures, StepStone Group, XTX Ventures and Zoom.

An enterprise can start from a capable open model and post-train with a fraction of the resources required at the frontier, making the model its own. Mercor says a growing number of companies see owning their intelligence as essential, and that a playbook is already visible across the industry: build public evals, post-train a model to cut costs, then train models for customers. Mercor says the capabilities it built for frontier labs are now being adopted by big tech, developers of open models, application-layer companies and the Fortune 500.

Professional work depends on two kinds of knowledge no model inherently has: how a task gets done, and how a specific company does it according to its systems, policies or quality bar; without both, an agent fails in production. The United States alone has more than 800 distinct occupations, each containing workflows that models have yet to be evaluated against. Mercor works with enterprises and experts in their fields to build custom evals based on the company's actual workflows; the results show how models perform against the company's requirements and help determine where additional training or practice in realistic environments could improve performance.

Mercor acquired Deeptune earlier this year, and its team spent two years recreating hundreds of enterprise applications where AI agents can practice professional workflows. Those recreated applications give agents a place to practice inside enterprise software environments rather than only on static evaluation items. Mercor says more than five million experts in its network are paid to train frontier AI models based on their unique knowledge, and that it has opened offices in New York and London to be closer to the enterprises and labs it serves.

The report covers the funding round, expert-network size, eval and post-training services and the Deeptune acquisition as Mercor announced them; it does not cover deployment results or model performance figures for any specific enterprise customer.

The material gives no specific metrics for the custom evals, no count of participating enterprises and no performance change after post-training; the projection of near double-digit GDP growth over five years is the company's extrapolation from its own ratio of inference spending to employee salaries, which it puts at 3x, and the material itself frames it as conditional.