Snorkel AI
Frontier AI data lab building expert-curated training datasets, evaluation benchmarks, runnable environments and specialized agents for AI teams.
At a Glance
About Snorkel AI
Snorkel AI is a frontier AI data lab, founded out of the Stanford AI Lab in 2019, that helps AI teams build the training data, evaluation systems and environments behind frontier models and agentic AI. Its offerings span expert data development, benchmarks and public leaderboards, and custom specialized agents. Work is delivered through a combination of platform, research and embedded services.
What It Is
Snorkel provides data development services for frontier labs and enterprises. It builds research-grade datasets, evaluation harnesses and runnable environments where generic coverage runs out. Delivered data includes expert demonstrations and reasoning traces, preference labels and rankings, and rubrics with verifiable outcomes. Environments include repo and CLI tools, browser/GUI harnesses, and multi-step stateful workflows.
How the Data Development Process Works
Snorkel describes its process around a few design decisions: well-specified expert-level tasks scoped to model failure modes, calibrated expert review, rubrics distilled into programmatic checks and evaluator models, and adjudication with full provenance trails. Edge-case coverage is expanded from expert-authored seeds across difficulty bands, and eval harnesses are built alongside the data to give reproducible scores across model versions.
Specialized Agents and Benchmarks
Snorkel says the same data development system powers its custom specialized agents, which are evaluated against task-specific rubrics and programmatic pass/fail checks. The company also publishes leaderboards across agentic coding, software engineering, computer use, knowledge work, science and enterprise environments, and runs an Open Benchmarks Grants program. Its research hub includes papers such as RIFT, OSWorld 2.0 and Agents' Last Exam.
Enterprise Platform and Security
The enterprise page describes Snorkel Flow, an integration-first platform with a Python SDK and no-code interfaces, deployable on private cloud, public cloud or Snorkel-hosted cloud and Kubernetes-native. Snorkel states it is SOC 2 Type II and HIPAA compliant, with SSO, role-based access controls, and encryption in transit and at rest.
Community Discussions
Be the first to start a conversation about Snorkel AI
Share your experience with Snorkel AI, ask questions, or help others learn from your insights.
Pricing
Contact sales
Snorkel Flow enterprise platform; pricing is not published and is available through sales.
- Deployment on private cloud, public cloud, or Snorkel Flow cloud
- Python SDK and no-code interfaces
- Snorkel Success Program
Capabilities
Key Features
- Expert-curated datasets for frontier AI
- Custom data development
- Expert demonstrations and reasoning traces
- Preference labels and rankings
- Rubrics and verifiable outcomes
- Standard and custom runnable environments
- Public benchmark leaderboards
- Open Benchmarks Grants
- Specialized agents
- Adjudication and provenance audit trails
- Python SDK and no-code interfaces (Snorkel Flow)
