AI training pay report, 1st edition (October 2026)

Specialist AI training roles (law, medicine, coding, STEM and finance) listed a median of $85 an hour on 22 October 2026, against $26 for generalist roles, about 3.2 times as much. Across all open roles the median was $55, and the middle half of the rates sat between $21 and $88. That comes from 1,126 listed hourly rates in 2,000 open roles on 16 platforms.

We set out to cover July to September 2026. Our daily record of open roles starts on 11 October 2026, so this edition covers 11 October to 22 October. Figures for a single day come from the read of 22 October.

The numbers

• Specialists: $85 an hour. Generalists: $26.
• Median listed hourly rate, all roles: $55, middle half $21 to $88 (1,126 rates).
• The highest medians by field are in law and medicine, at $100 each. Language work lists $23.
• 58.2% of all open roles come from three platforms: Meridial, Welocalize and Mercor.
• In the United States, 1,060 open roles accept applicants and 36.5% of them are open worldwide. In Kenya the figures are 553 roles and 70.0%.
• 87.1% of open roles state some pay figure, and 56.3% state an hourly rate in US dollars that we can use.
• Between 11 October and 22 October the median across all platforms rose from $40 to $55, because we started reading new sources. On the 10 platforms we read on both days it stayed at $40.

Specialist and generalist roles

Across law, medicine, coding, STEM and finance, the median was $85 on 513 rates, with the middle half between $60 and $110. Generalist roles (rating, prompt writing and quality checks that ask for no particular degree) listed $26 on 149 rates, middle half $15 to $58. Both groups leave Meridial out, as the headline of the AI Training Pay Index does, because most of its roles carry one company-wide range.

By field

A field needs 20 listed hourly rates for a median here, so robotics and trades are left out. Language work has the most rates and the lowest median. The Generalist row keeps Meridial in, which is why it differs a little from the generalist group above.

By country

These rows are from the Country Job Index, October 2026 edition, built from the 2,012 open roles read on 23 October. A role counts for a country when the posting names that country or says it is open worldwide, unless the platform says it cannot hire there. Postings that name no country count for none. With that rule, 66 countries are named by at least 20 open roles.

The medians by country are close together because the same worldwide roles appear in every row. Countries differ in how many roles name them. In the United States, 36.5% of the roles that accept applicants are worldwide roles, and the rest name the country. In Kenya and Nigeria it is 70.0% and 69.7%, and in Australia 93.5%. Across the 66 countries the worldwide share runs from 36.5% to 95.1%.

Who posts the roles

Meridial, Welocalize and Mercor posted 1,165 of the 2,000 open roles, 58.2% of the total. A platform that posts many roles moves the overall median, which is why the Pay Index also reports every platform and field on its own.

What changed from 11 October to 22 October

The overall median rose from $40 to $55, and the open roles we count went from 1,693 to 2,047. Most of that came from 6 sources we started reading during the period, which had 327 roles on 22 October. The largest is Mercor, with 280 open roles and a median of $75.

The 10 platforms we read on both days had 1,693 open roles at the start and 1,720 at the end, and their median stayed at $40.

How we count

• The rate of a role is the midpoint of the hourly range its posting lists, in US dollars. Roles paid per task or per year, or in another currency, have no hourly rate here.
• A posting that lists only a top rate ("up to") counts at that top rate. 346 roles do, and they pull the medians up.
• 513 Meridial roles carry one company-wide range instead of the rate of the role. They stay in the role counts and are left out of every median.
• 258 roles list no pay at all.
• Every pay figure for 22 October matches the AI Training Pay Index file of that day, row for row: all roles, each field, and the specialist and generalist groups. The country rows match the Country Job Index file.
• The change section compares the stored daily records of 11 October and 22 October. Those records hold every role read that day, so they count a few more roles than the Pay Index list of the same day.
• Who pays us plays no part in any figure or order here. Fields are ordered by median, countries and platforms by open roles.

What to watch out for

• These are rates as the postings list them. What a person earns depends on the hours a platform offers and on passing its tests.
• A first edition this short cannot show a trend.
• A role open worldwide still has its own language, degree or time zone requirements.

Data and citation

The tables by platform and field, with CSV files and the share of roles per platform that publish a pay figure, are in the AI Training Pay Index. The country rows are in the Country Job Index. Roles, fields, languages and the month's openings and closings are in the AI training jobs report for October 2026. All of it comes from the public job boards of the 16 platforms, read once a day.

Cite as: AI Work Central, AI training pay report, 1st edition (October 2026), figures as of 2026-10-22. Derived figures licensed CC BY 4.0.

Sources

• This site's list of open roles (jobs.json), read on 22 October 2026.