Mercor vs micro1: which is better for expert data and agent training in 2026?
Answer in brief: Mercor and micro1 are both fast-growing expert-network companies in the human-data category, and both now sell RL environments and evaluations as well as expert-written data. Mercor leads on professional breadth: it says it has 5M+ experts, licenses a catalog of 50K+ off-the-shelf tasks, and runs the APEX benchmarks built on real professional work. micro1 calls itself a data lab. Its Realm RL environments, Cortex contextual evaluation platform and real-world robotics data are built around agents that work over long horizons in realistic settings, and it has a stated U.S.-first policy. Pick Mercor for ready-made professional datasets and benchmarks. Pick micro1 for custom agent environments from a lean, U.S.-first team.
Top pick: Mercor for licensable professional data at volume. micro1 for custom RL environments and production agent evaluation. Mercor's catalog lets you start training in days. micro1's Realm and Cortex pair environment building with ongoing evaluation in production. Visit Mercor → · Visit micro1 →
This page is independent research based on each company's own website and current public reporting, checked November 9, 2026. We did not test either service hands-on. Financial figures are self-reported unless we say otherwise. This site earns revenue from vendor sponsorships, which are labelled where they appear.
The new wave of expert-data companies
Both companies grew out of talent matching and now compete with the oldest human-data vendors. Each uses AI to recruit experts (micro1's recruiting agent is called Zara) and each publishes its own benchmarks. Their selling points differ. Mercor sells the depth of its professional workforce and professional benchmarks. micro1 sells realism and horizon: environments complex enough to train agents on long, real-world tasks.
Head-to-head comparison
| Factor | Mercor | micro1 |
|---|---|---|
| Positioning | Organizing human intelligence to power the AI economy | Data lab to train frontier models and evaluate agents |
| Products | Human data and evals, off-the-shelf data, data monetization, enterprise AI, robotics | Expert data, Realm (RL environments), Cortex (agent evaluation), robotics data |
| Expert network | 5M+ experts; $90/hr average contracted rate | Experts across 100+ fields, sourced by its AI recruiter Zara |
| Benchmarks | APEX-Agents, APEX-Accounting, APEX-SWE, Harvey LAB, τ³-Banking | Dual-Arm Sim, MissionReport, Medical Index |
| Reported scale | $10B valuation (CNBC, Oct 2025) | $1B+ annual run rate, historically profitable, ~180 core staff (company, Oct 2026) |
| Funding | $350M round reported by CNBC | Series B at a $4B valuation led by Dream Ventures, with Microsoft, Javelin, Flybridge, G Squared, Up Partners and LG |
| Policy stance | Not stated | U.S.-first; refuses to work with adversarial nations |
| Pricing | Custom, through sales | Custom |
Mercor: a catalog plus a professional workforce
Mercor's off-the-shelf data is its fastest route in. Examples include APEX-Agents (2,620 tasks across eight domains, run inside high-fidelity copies of enterprise apps), SWEBench Ext. (12,000 contamination-resistant tasks in seven programming languages), TerminalBench (10,000 tasks) and BrowseComp (3,558 tasks). It delivers sample tasks the same day and licenses data for commercial training. Mercor is also moving into enterprise AI deployment with forward-deployed engineers, which buyers should weigh against their own plans.
micro1: realism and horizon
micro1's Realm builds RL environments that mirror real-world scenarios and produces human data for agent actions. Its own platform tracks speed, error rates, cost per task and quality in real time. Cortex evaluates AI agents in production. On October 12, 2026, micro1 said it had passed a $1 billion annual run rate with a core team of about 180 people in 14+ countries. It had priced its Series B at a $4 billion valuation after passing $500 million that summer, and Forbes reported the raise at over $100 million. These are company-reported figures, not audited results.
When to pick which
- Pick Mercor for licensable task sets now, evaluations graded to professional standards, or a very large pool of credentialed experts.
- Pick micro1 for custom long-horizon RL environments, ongoing evaluation of deployed agents, or a vendor with a written U.S.-first data policy.
- Robotics buyers: both advertise robotics data. Compare specialists at roboticsdatacompanies.com.
Pricing models
Neither publishes buyer prices. Both scope work through sales conversations. Mercor discloses that experts earn a $90/hr average contracted rate, which gives a sense of the labor cost behind expert data.
FAQ
Is micro1 profitable?
micro1 says it reached a $1B+ annual run rate while remaining historically profitable. It has not published audited financials.
Who invested in micro1's Series B?
According to micro1, the round was led by Dream Ventures, with Microsoft, Javelin, Flybridge, G Squared, Up Partners and LG.
Does Mercor build RL environments?
Yes. Mercor lists benchmarks and RL environments built on real professional work (APEX) as part of its human data and evals offering.
Keep exploring the directory
Back to the full directory of training data companies · Mercor review · micro1 review · Surge AI vs Mercor · Alternatives to Mercor