Snorkel AI
AI training-data and evaluation specialist building research-grade datasets, agents, and benchmarks for frontier AI labs.
About
Snorkel AI builds specialized training data, evaluation systems, and runnable environments for frontier AI models. The work covers expert-curated datasets, custom evals for specific failure surfaces, and bespoke agents scored against task-specific rubrics. Public benchmarks include Terminal-Bench 2.0, SlopCodeBench, and an Agentic Coding benchmark.
Founded out of research on programmatic labeling and weak supervision. Stated mission is to make AI data development programmatic, like any other type of software development. Technology has been developed and deployed with Google, Apple, DARPA, and Stanford Medicine. Research team has produced 170+ peer-reviewed publications.
Raised $100M in May 2025 from Greylock, Google Ventures, Lightspeed, IQT, Addition, BlackRock, SV Angel, and Walden Catalyst. Customers described as world-leading enterprises across private and public sectors, with systems processing billions of queries and records.