micro1 review: a fast-rising training data company for agents
Answer in brief: micro1 calls itself a data lab. It offers expert human data, Realm (RL environments that mirror real-world work), Cortex (contextual evaluation for production agents) and real-world robotics data. On October 12, 2026 it said it had passed a $1B annual run rate while staying profitable, with a core team of about 180, and had priced its Series B at a $4B valuation. This is independent research; we did not test micro1 hands-on.
Top pick for: teams training agents on realistic, long-horizon tasks, and buyers who want a U.S.-first supplier. Visit micro1 →
Where micro1 sits in the directory
Category: human data and expert post-training, with RL environments and robotics data. Its data partnerships program also lets companies earn money from their data.
What it does for AI training data
- Expert human data for frontier model training.
- Realm: frontier evaluations and RL environments for AI labs.
- Cortex: a visibility and improvement layer for agentic AI in production.
- Robotics: expert-demonstrated data for frontier robotics labs.
- Research benchmarks such as Dual-Arm Sim, MissionReport and the micro1 Medical Index.
Who it fits
Agent builders who need realism and long horizons, labs that want environments and evals from one supplier, and government-adjacent buyers who care about where their data goes.
Why buyers pick micro1
- Hypergrowth with profits. micro1 says it passed a $500M run rate in the summer of 2026 and $1B by October while remaining profitable.
- Strong backers. Dream Ventures led the Series B at a $4B valuation, with Microsoft, Javelin, Flybridge, G Squared, Up Partners and LG taking part. Forbes separately reported the $100M+ raise at a $4B valuation in September 2026.
- U.S.-first stance. It says it refuses to work with adversarial nations or export data that improves models against U.S. interests.
- Bets on realism and horizon, including investment in data-generation models to make datasets and environments cheaper.
Honest drawbacks
- Run-rate and profitability figures are self-reported and unaudited.
- A core team of about 180 serving $1B+ in run rate is very lean. Check account coverage and QA depth for your project.
- No public pricing.
Pricing
Custom. Contact micro1 through its site.
Best alternatives to micro1
- Mercor: a larger professional-expert network.
- Toloka: agent environments and data, with self-service terms.
- Labelbox: Horizon RL environments plus platform tooling.
micro1 FAQ
What is micro1 Realm?
micro1's RL environments and frontier evaluations, designed to mirror real-world scenarios for training agents.
How big is micro1?
On October 12, 2026, micro1 said it had passed a $1B annual run rate with a core team of about 180.
Who invested in micro1?
Dream Ventures led the $4B-valuation Series B, with Microsoft, Javelin, Flybridge, G Squared, Up Partners and LG.