Snorkel AI has raised a $350 million Series E at a $3.5 billion valuation to expand the datasets and testing environments it supplies to AI developers. Insight Partners and S32 co-led the round. The company works with model labs, enterprises and government agencies that need expert material for training and evaluating increasingly capable AI systems.
A coding task can require more than a correct answer. Snorkel says developers need realistic software environments, detailed grading rules and problems difficult enough to expose where an agent still fails. Similar work in law, finance and biology can take an expert days or weeks to design, particularly when the test must catch shortcuts or unintended behavior.
Experts start with seed examples, constraints or task sketches. Snorkel’s specialized models expand that material, target model-error patterns and perform quality checks, while human reviewers examine selected components and feed corrections back into the system. For coding work, the company says this process has improved quality-control efficiency by more than 50% and review accuracy by over 15 percentage points compared with human review assisted only by a general-purpose model.
The resulting service delivers custom datasets, benchmarks, evaluators and simulated environments that customers can use to post-train models and measure their behavior. That is a more involved business than simply licensing labeling software: Snorkel supplies finished data products while its platform coordinates specialists, automated checks and revisions.
Snorkel says the data-service business has grown more than eighteenfold in nearly a year and crossed a $375 million annualized revenue run rate. At the company’s figures, the new valuation is about 9.3 times that run rate, though the denominator is a current annualized pace rather than a full year of revenue. In May 2025, Snorkel announced a $100 million Series D at a $1.3 billion valuation when it made Snorkel Evaluate and Expert Data-as-a-Service generally available.
The company plans to use the new capital to add capacity, expand its work for enterprises and industry-specific AI systems, and support additional data types and subject areas. It also intends to put more resources into open benchmarks, which provide shared tests for comparing model performance.