micro1

micro1 is a data lab supplying expert human data, reinforcement learning environments and contextual agent evaluation to teams building frontier AI, backed by public model benchmarks and an expert hiring pipeline.

micro1 positions itself as a data research company for teams that build and test frontier AI models. Rather than a self-serve app, it offers infrastructure built around three ideas: expert human data, training environments that mimic real-world work, and evaluations that take context into account. The homepage groups the offering into three products: Realm, Cortex and Robotics. Realm provides reinforcement learning environments that mirror real scenarios, with the goal of producing human data for agentic actions and stronger model reasoning. Cortex is described as a contextual evaluation platform for improving AI agent performance once agents are running in production. Robotics covers high-fidelity real-world data for training embodied systems. Together they span pre-deployment training, post-deployment measurement and physical-world data collection. The site also carries a public research side. Benchmarks such as Dual-Arm Sim (careful physical work in sensitive settings), MissionReport (government status reporting) and a Medical Index that combines the company's medical benchmarks all publish ranked leaderboards of frontier models. These pages work as both proof of method and a reference for anyone comparing model families on specialised tasks. A stated research aim concerns deciding where each person should spend their time. In practice the audience is split in two. Companies and AI labs use the contact form to describe a project and start a data partnership conversation. Domain experts use a separate opportunities page to apply for roles that supply the human judgment behind the datasets. The model resembles a managed data vendor with a marketplace of experts attached, not a software subscription. Among alternatives, it sits closest to human data and evaluation providers that supply labelled data and expert review to model builders, and further from lightweight tools for tracking prompts or logging outputs. Teams wanting a quick, self-serve eval dashboard may find the engagement style heavier than they need, while labs that need specialist annotators and custom environments are the natural fit.