Open Data

The State of the AI Training Job Market

Name: AI Training Job Market Statistics
Creator: aitrainer.work
License: https://creativecommons.org/licenses/by/4.0/

Pay, volume, and hiring terms compiled from 4,433 listings across 13 platforms, tracked since 2025-12-30. Free to reuse under CC BY 4.0.

Last updated: July 2026

$55/hr

median rate (n=4,208)

25.9 days

median time a listing stays open

93%

of disclosed listings require an interview

Median hourly pay by field

USD, hourly-priced listings only

Legal

$100

n=127

Healthcare

$90

n=170

Accounting

$83.8

n=50

Finance

$80

n=138

Business

$75

n=403

STEM

$70

n=324

Design

$70

n=39

Humanities

$70

n=41

Data Science

$57.5

n=138

Mathematics

$56.8

n=46

Median hourly pay by platform

USD, hourly-priced listings only

Ethos

$100

n=136

Handshake

$90

n=100

Mercor

$80

n=1239

Mindrift

$55

n=52

Vetto

$52

n=17

SME Careers

$50

n=284

Micro1

$47.5

n=1593

Alignerr

$45

n=249

Turing

$40

n=484

Outlier

$25

n=38

Ethos: Ethos is an expert network sourcing senior/PhD-level consulting placements for undisclosed third-party clients, not a standard crowdsourced AI-training platform — its pay figures reflect that different labor market rather than a directly comparable "platform rate."

Listing volume by field

Total listings observed, all platforms

Generalist

502

n=502

Software Engineering

463

n=463

Languages

438

n=438

Business

429

n=429

STEM

358

n=358

Healthcare

177

n=177

Data Science

163

n=163

Finance

151

n=151

Legal

135

n=135

Writing

104

n=104

New listings per month

Excludes each platform's onboarding backfill

Every figure carries a sample size (n). Breakdowns with fewer than 10 listings are omitted rather than shown as noise. See methodology below.

Download the full dataset

A static, cached JSON file with everything on this page plus pay-by-month, contract duration, and weekly-hours breakdowns. No API key, no rate limit, no auth required. It's just a plain, static file rebuilt on every deploy.

https://aitrainer.work/data/market-trends.json Download ↓

Citing a specific figure? Use the dated snapshot instead so the numbers don't move under you: https://aitrainer.work/data/market-trends/market-trends-2026-07.json

License: CC BY 4.0

This data is free to use in your articles, research, or apps, including commercially. You just have to link back to this page. That's the whole deal.

Copy this to attribute us

Data from <a href="https://aitrainer.work/open-data">AI Trainer Jobs</a>, licensed under <a href="https://creativecommons.org/licenses/by/4.0/">CC BY 4.0</a>.

Data from AI Trainer Jobs (aitrainer.work/open-data), licensed under CC BY 4.0 (creativecommons.org/licenses/by/4.0)

Full license: creativecommons.org/licenses/by/4.0

Methodology & limitations

What's in the corpus: Every listing observed since tracking began, whether or not it is still open today. Listings are not split into "active" and "expired" because that split reflects when this file was generated, not the market.
"First seen" dates: The date this project first observed a listing, not the date the platform published it. A listing posted before tracking began on that platform is dated to the day it was first seen. The first month of coverage for each platform contains a backlog of listings that were already open. Months where this applies are marked includes_backfill: true and should be excluded from growth comparisons.
Pay figures: Hourly-priced listings only, in USD. Where a listing advertises a range, the midpoint is used. Rates below $3/hr or above $400/hr are treated as parsing errors and excluded. Per-task, monthly, and annual listings are excluded from pay figures.
Fields: Field (domain) is assigned from the listing title, description, and platform tags. Roughly a third of listings carry no confident field assignment and are excluded from field breakdowns.
Listing lifespan: Days between first and last observation of a listing. Listings still open at generation time have no end date and are excluded, which biases this figure slightly downward.
Known limitations: Coverage is not the whole market: it is the platforms this project tracks. Platforms differ in how completely they disclose pay, so per-platform sample sizes vary widely. Always read median figures alongside sample_size.