Mercor
Monthly visits
14,164,192
2026-10
Monthly growth +20.4%

Mercor - Expert-built evaluations and human data for AI systems

Mercor connects enterprises and AI labs with professional experts, evaluation suites, benchmarks, and human data services. Its evaluations test agents against real workflows using expert-built tasks and scoring rubrics. Enterprises can also map workflows, deploy AI systems with embedded engineers and integrations, and monetize business workflow data with controls over what is shared.

Visit WebsiteUpdate this product

0Views

Capabilities

Expert-built evaluations

Creates repeatable tests and scoring rubrics to measure an AI agent’s performance and operating cost. Experts turn customer workflows into tasks, then score agents against criteria; evaluations can run on deployed agents and real traffic or offline in simulated worlds.

APEX benchmarks

Provides benchmark tasks and leaderboards for assessing AI models on professional work. The benchmarks cover domains including accounting, software engineering, law, and medicine, and may use real tools, professional task artifacts, and expert-authored scoring.

Human data and expert network

Sources professional experts and supports human evaluation, AI training, and expert data work. Enterprises can engage experts for evaluations and post-training, while data services include automated masking of more than 60 categories of sensitive identifiers and organization control over shared data.

Enterprise AI deployment

Supports identifying AI opportunities and deploying production AI for enterprise workflows. Mercor maps workflows and institutional knowledge, embeds Forward Deployed Engineers, and provides enterprise integrations plus tools for evaluation and ongoing optimization.

Pricing

Mercor’s supplied pages do not publish plan prices or a standard plan list. Enterprise evaluation work can be scoped through a conversation, and task-based evaluation pricing is described as priced per task rather than per expert headcount.

Use cases

Evaluate an AI agent on real work

When an enterprise needs evidence about whether an agent works on its own workload, Mercor builds a repeatable suite of expert-authored tasks and rubrics. Teams can run evaluations against deployed agents and real traffic or compare models in offline simulated benchmarks, and inspect quality and cost.

Choose a model and agent configuration

When comparing models or agent configurations for a workload, teams use evaluation results and the price–performance frontier. The results help identify which model clears a chosen quality bar at the lowest cost and which tools or context improve the agent.

Monetize business workflow data

When an organization wants to create revenue from data generated by its existing workflows, Mercor offers data monetization services. Automated anonymization masks 60-plus categories of sensitive identifiers, and the organization retains control over what data is shared.

Limitations

  • The disclosed evaluation delivery period is four to six weeks from the first conversation to a reusable suite; the exact timing depends on the engagement.

  • Production Rubrics evaluate a deployed agent using live production traffic; APEX benchmarks run offline in expert-built simulated worlds.

  • Mercor’s data monetization service gives organizations control over what data is shared, but the supplied pages do not specify the data license terms or payout amounts.

Who it is for

  • Enterprise teams building or evaluating production AI agents

  • AI research labs seeking professional human data, evaluation, or post-training support

  • Professional experts seeking paid AI evaluation, training, or data work

FAQ

What does an evaluation measure?

An evaluation tests what an AI agent can do on a particular workload using a repeatable set of tasks, rubrics, and scoring. Mercor says results can inform whether the agent works, which model or configuration to use, and what each option costs.

Can Mercor measure the cost of running an agent?

Yes. Mercor describes a price–performance frontier for the customer’s workload, which can help select the least costly model that meets a quality bar and identify appropriate reasoning and context settings.

How do Production Rubrics differ from APEX benchmarks?

Production Rubrics run live against a deployed agent and real traffic. APEX benchmarks run offline in expert-built simulated worlds as a fixed, repeatable set for comparing models and configurations before deployment.

How long does it take to build an evaluation suite?

Mercor says a reusable evaluation suite takes four to six weeks from the first conversation. The enterprise evaluation page also describes delivery in less than six weeks.

Who owns the evaluation suite after delivery?

The customer keeps a reusable suite built for its team, rather than only receiving a one-time report.

Can Mercor create a custom benchmark for an agent?

Yes. Mercor says APEX benchmarks can be purpose-built for a domain, using expert-authored simulated worlds and tasks that mirror the agent’s work.

Comments

Please sign in to leave a comment.
No comments yet. Be the first to share your thoughts!