Micro1

Micro1 is launching flow, a data platform aimed at frontier AI labs that need training data and evaluation environments capable of pushing model capabilities beyond what public datasets can support. The system centers on Realms, real-world reinforcement learning environments where domain experts define what strong performance looks like on a given task, then layers on FlowGen to expand those seeds into full rubrics, edge cases, and variations, and FlowQC to grade model attempts against expert criteria and route the highest value cases back for human review. A companion product, Cortex, is positioned for enterprises that want to diagnose where their deployed agents fail and feed those findings back into targeted data collection. The launch matters because the bottleneck in model improvement has shifted from compute and algorithms to the supply of high horizon, high realism data, and the major labs are increasingly buying that data as a service rather than building it in house. Micro1 was founded by Ali Ansari, who leads the company as CEO and has grown it into an AI human data training platform that partners with frontier AI labs to deliver expert-led data and evaluation at scale. The company has also drawn attention this year for reaching a $500M gross run rate amid the AI training boom , and it has been publishing its own Realm benchmark series scoring frontier models on expert workflows in fields like law. Flow is framed as a continuous loop rather than a one-off dataset vendor relationship, with the platform recursively improving as it ships what Micro1 calls units of intelligence improvement back to model providers. The roadmap extends the same architecture into physical world data for robotics, combining human demonstration with multimodal quality checks to build sensor fused datasets for world models and control policies, which puts Micro1 in direct competition with the incumbent human data providers serving both LLM labs and the newer wave of robotics foundation model teams.