How much do AI training data providers charge in 2026?
Last updated: November 9, 2026
Most AI training data providers do not publish prices. They quote per project, based on task difficulty, expertise required, volume, turnaround and quality targets. Prolific is the main exception with a public fee schedule, and Surge AI offers off-the-shelf datasets you pay for only if they improve your metrics. Everything below comes from the providers' own pages as of November 9, 2026. We did not obtain quotes.
Pricing models at a glance
| Provider | Pricing model | What is public |
|---|---|---|
| Prolific | Pay as you go, or managed under a statement of work | Participant reward you set plus a platform fee: usually 42.8% for corporate, 33.3% for academic or non-profit customers. Minimum reward $8/hr; $12/hr recommended. No contract or subscription for the platform. |
| Surge AI | Off-the-shelf datasets plus custom work | For off-the-shelf data: train on the full dataset first and pay only if it improves your metrics. Custom work is quoted. |
| Mercor | Custom engagements | Lists a $90/hr average contracted rate for experts (what experts earn, not the buyer price). |
| Handshake AI | Custom, via a scoping session | Expert roles posted at rates up to $150/hr for software engineers and up to $17/hr for generalist AI evaluation specialists (expert pay, not the buyer price). |
| Toloka | Self-service platform plus managed projects | Self-service agreement is published; managed pricing is quoted. |
| Scale AI | Demo-led contracts | None. |
| Labelbox | Sales-led | Pricing page returned 404 on 2026-11-09. |
| micro1, Turing, Snorkel AI, Invisible, iMerit, Appen, Sama | Custom quotes; several offer samples or pilots | None; Turing and Snorkel invite dataset sample requests. |
What drives the price
- Expertise: data from credentialed experts (physicians, attorneys, senior engineers) costs far more than generalist labeling. Posted expert rates show how wide the range is.
- Task complexity: multi-step agent trajectories, RL environments and long-horizon rubrics take more time than simple classification.
- Quality assurance: multi-stage review, gold sets and calibration add cost but reduce rework.
- Modality: LiDAR, 3D sensor fusion, video and medical imaging need specialized tooling and trained teams.
- Exclusivity: custom, exclusive data costs more than licensing an off-the-shelf dataset.
How to get a fair quote
- Write a precise task specification with examples and acceptance criteria.
- Ask at least two or three providers for samples or a small paid pilot.
- Compare cost per accepted item and measured model lift, not the hourly or per-task rate alone.
- Confirm who owns the data, whether it is exclusive, and how your inputs are secured.
For the full ranking and each provider's strengths and drawbacks, see the main guide. For how we researched it, see our methodology.