A Guth Labs publication ↗
Back to News

Mercor raises $750 million as it brings AI evaluation and training into enterprise workflows

A human hand calibrating a constellation of work artifacts, an AI-generated illustration of AI evaluation.
AI-generated editorial illustration for guthlabs.newsroom.

guthlabs.newsroom | Enterprise AI | November 10, 2026
Mercor announced a $750 million Series D at a $20 billion valuation on November 10, backing an expansion from work with frontier AI laboratories into enterprise evaluation, post-training and agent deployment. The company said participants included Benchmark, Felicis, NVIDIA and funds managed by affiliates of Apollo, among other investors. Mercor Series D announcement

The round also included an employee tender. Mercor did not break out the tender’s size in its announcement, so the headline financing figure should not automatically be treated as the amount of new operating cash available to the company. The stated uses of funding include evaluation research, post-training infrastructure and additional environments in which agents can practice complicated work. Mercor Series D announcement

Mercor says it is applying expertise developed with model laboratories to companies that need to determine whether an AI system can perform their own tasks. The distinction is important: a model may demonstrate broad professional knowledge while failing on a particular organization’s systems, approval rules or definition of a successful result. Mercor Series D announcement

Measuring a job rather than an answer

The company’s existing APEX-Agents benchmark offers one example of that approach. It covers multi-step work associated with investment banking, management consulting and corporate law, using 240 tasks across 31 constructed project environments. Mercor says professionals designed the scenarios, tasks and grading criteria, and that the dataset and execution infrastructure are available publicly. This is background to its evaluation business, not a benchmark newly announced with the financing. Mercor APEX-Agents description

The design puts the emphasis on completing work across applications. That is a more demanding question than whether a model can answer a related exam problem. A professional deliverable can require collecting evidence, choosing a method, manipulating files and checking whether the final result satisfies the original request.

Mercor’s enterprise offering includes identifying candidate workflows, building agents around customer systems and reviewing existing deployments for improvements. The company describes work on model selection, prompts, tools, architecture and monitoring. It also offers data licensing as a separate, opt-in service under which customers choose what information may be made available. Mercor enterprise offering

The value depends on the evaluation

A customer-specific test can be valuable because it makes success explicit. But the evaluation itself must be examined. Tasks should represent the difficult and uncommon cases that matter to the business, rather than only examples on which automation already performs well. Reviewers also need to distinguish a plausible-looking output from a correct, authorized and complete result.

There is a second issue when the same provider helps train, deploy and evaluate a system. Those roles can create a productive improvement cycle, but customers still need tests that were not used to tune the agent. Held-out work, clear scoring standards and independent review can help show whether gains extend beyond familiar examples.

The Series D gives Mercor more resources to build around that cycle. Its valuation signals investor expectations, while the tender provides a separate liquidity component. For enterprise buyers, the useful measure will be whether evaluation-led deployment reduces failures and operating costs on real work. More extensive training infrastructure is an input to that result, not proof that it has already been achieved.

Sources and citations

Revision history

Published .