Snorkel AI / Snorkel Flow

Snorkel AI

by Snorkel AI, Inc. (CEO and co-founder Alex Ratner)

Started as a programmatic-labeling / weak-supervision platform (Snorkel Flow); by 2026 it sells expert-curated data, evaluations and environments to frontier AI labs and enterprises.

2026 position (read this first)

• 2026-09-22: $350MSeriesEata$3.5B valuation, led by Insight Partners and S32; reported annualized revenue run rate **$375M,up18xin12months\ast \ast (TechCrunch,companypressrelease).Previousround:$100M Series D in May 2025 at $1.3B, which introduced Snorkel Evaluate and the expert Data-as-a-Service offering.
• Business-model shift: TechCrunch describes Snorkel moving from data-labeling automation software to delivering finished datasets (human experts plus synthetic data). Snorkel’s homepage now lists three offerings: Data Development (expert-curated datasets for frontier AI), Specialized Agents, and Research and Benchmarking (leaderboards). The homepage does not mention Snorkel Flow, so its current status as a standalone product is unconfirmed.
• Competes for AI-lab data spend with Mercor, Handshake and Micro1 (TechCrunch).
• Origin: Stanford AI Lab research on weak supervision (started ~2015), company founded 2019; the open-source snorkel project came first. Alex Ratner is also an affiliate assistant professor at the University of Washington.

How the original technology works

• Labeling functions: experts write small rules/heuristics/model calls that emit noisy labels or abstain.
• A label model estimates each source’s accuracy and correlations from agreements and disagreements and outputs probabilistic training labels; a downstream model is trained on them.
• Iterate on labeling functions against a small gold dev set. Trade-off: needs domain expertise, labeling functions can encode bias, and gold labels are still needed for evaluation.
• Research basis: Ratner et al., “Snorkel: Rapid Training Data Creation with Weak Supervision” (arXiv 1711.10160).

Plans

Pricing not public; enterprise sales and services.

Related

agent-evals-and-observability (evaluation tooling), ai-api-platforms.

Open items

• Snorkel Flow’s current availability and the open-source library’s maintenance status are not confirmed (not on the homepage; not checked on GitHub). Investor lists differ between reports (Greylock vs Lightspeed/GV as existing participants).
• Removed as unverified: 2025 customer list (BNY Mellon, Memorial Sloan Kettering, Chubb, Wayfair, Experian) and generic pilot advice.

Sources

• https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/ (accessed 2026-10-02)
• https://www.prnewswire.com/news-releases/snorkel-ai-raises-350m-to-scale-the-data-factory-for-frontier-ai-302886796.html (search result, accessed 2026-10-02)
• https://snorkel.ai (accessed 2026-10-02)
• https://arxiv.org/abs/1711.10160 (accessed 2026-10-02)