Snorkel AI

Snorkel AI supplies finished training datasets, reinforcement-learning environments and evaluation harnesses to frontier AI labs, hyperscalers, enterprises and government agencies. It is for teams that need expert-quality data to train or test specialized models and agents and cannot staff the domain experts and quality controls themselves.

Snorkel AI began as a Stanford AI Lab research project on programmatic data labeling and launched commercially in 2019 with Snorkel Flow, software that lets data scientists write labeling functions instead of hand-labeling every example. Since September 2025 most of its business has been data-as-a-service: it delivers finished training and evaluation datasets, reinforcement-learning environments and benchmark harnesses to frontier AI labs, hyperscalers, enterprises and U.S. government agencies. The work combines Snorkel's own software and models, which generate data synthetically, with a network of paid subject-matter experts in fields such as coding, law and medicine. Reviewers are calibrated against gold sets, and outputs go through rubrics, programmatic checks, fine-tuned evaluator models and multi-reviewer adjudication with an audit trail. Snorkel also builds public benchmarks, including Terminal-Bench, SWE-bench CLI and Senior SWE-Bench, and funds outside benchmark work through a $3 million Open Benchmarks Grants program. On September 22, 2026 it announced a $350 million Series E at a $3.5 billion valuation, led by Insight Partners and S32, nearly triple the $1.3 billion valuation of its May 2025 Series D. TechCrunch puts its annualized revenue run rate at $375 million, an eighteenfold rise in a year, while Snorkel's own release says more than $350 million. Expert payments are booked as cost of goods sold, so the figure is not directly comparable to the gross marketplace revenue reported by Mercor, Handshake and Micro1. No customers are named on the record for the data business.

Category

Data & Analytics

Pricing

Contact for pricing

Target Market

Heads of AI, ML Engineers, Data Scientists, Frontier AI Labs, Government Agencies

Deployment

Cloud-first

Founded

2019

Headquarters

San Francisco, USA

• ✓
  Expert data-as-a-service

  Delivers finished, curriculum-structured datasets with rubrics, reviewer guidance and difficulty tiers, so buyers receive training-ready data rather than labeling tooling.

• ✓
  Reinforcement-learning environments

  Standard and custom environments including repo and CLI tools, browser and GUI harnesses and simulated systems, used to train and test agents on realistic tasks.

• ✓
  Evaluation harnesses and benchmarks

  Task-specific rubrics, deterministic graders and reproducible scoring, plus public leaderboards such as Terminal-Bench, OSWorld 2.0 and SnorkelFinance 2.0.

• ✓
  Calibrated expert network

  Paid specialists in coding, law, medicine and other fields are trained against gold sets, which keeps expert judgments consistent across large projects.

• ✓
  Quality control with provenance

  Programmatic checks, fine-tuned evaluator models and multi-reviewer adjudication with full audit trails show how every data point was produced and approved.

• ✓
  Snorkel Flow programmatic labeling

  The original software lets data scientists write labeling functions that auto-label large datasets, with lineage tracking for auditing and updating rules.

• •
  Post-training a coding agent

  A model team buys expert coding demonstrations and repo-based RL environments with tasks that would take experienced developers days, then measures gains on held-out benchmarks.

• •
  Pre-deployment agent evaluation

  An enterprise builds a rubric-graded eval harness for its own workflow so it can score candidate agents reproducibly before rollout.

• •
  Insurance claims and underwriting data

  QBE's North American teams used Snorkel Flow to prepare company data for machine learning in claims and underwriting use cases.

• •
  Auto-labeling a large catalog

  A data science team writes a handful of labeling functions to label an entire content catalog's metadata, then audits the rules through lineage tracking.

Snorkel publishes no list prices. Data, environments and evaluations are sold as custom engagements through sales (the site offers dataset samples on request), and a G2 reviewer describes Snorkel Flow as expensive, enterprise-only and without a self-serve tier. Expert payments are booked as cost of goods sold, so budget for both the data and the expert time behind it.

This page was written from 9 sources, 7 on domains other than snorkel.ai.

1. 1.snorkel.ai — snorkel.aivendor
2. 2.snorkel.ai — snorkel ai valued at 3 5 billion amid surging demand for comvendor
3. 3.security.snorkel.ai — security.snorkel.ai
4. 4.techcrunch.com — snorkel ai triples valuation to 3 5b as demand for ai traini
5. 5.implicator.ai — snorkel ai 350m series e
6. 6.hedgeco.net — snorkel ai raised 350 million at a 3 5 billion valuation
7. 7.techstrong.ai — snorkel ai secures 350m for more complex ai training and eva
8. 8.coverager.com — snorkel ai raises 350 million
9. 9.hn.algolia.com — search