What Snorkel AI does
Snorkel AI helps frontier labs and enterprise AI teams develop specialized training data and evaluation environments to differentiate their models and agents. It focuses on building research-grade datasets and benchmarks for cases where generic, off-the-shelf data falls short.
Key capabilities
The company creates expert-curated datasets tailored to specific model failure modes, using calibrated expert review, task-specific rubrics, programmatic evaluation, and multi-reviewer adjudication pipelines with full audit trails. It builds custom AI agents grounded in domain-specific data and evaluates them against task-specific criteria rather than generic benchmarks. Snorkel also publishes peer-reviewed research, operates public benchmarks such as Terminal-Bench 2.0 and an agentic coding benchmark, and funds open-source AI research through Open Benchmarks Grants.
Who it's for
Snorkel AI serves frontier AI labs, academic research institutions, and enterprise teams deploying AI agents in high-stakes domains such as legal, insurance, and coding. The company originated from the Stanford AI Lab and emphasizes research-driven data quality and evaluation rigor.