The median advertised hourly rate was $68 per hour in October 2026, measured part way through the month. A quarter of the listings advertise $45 or less. A quarter advertise $93 or more. These are rates written in job listings, not what any person earns.
Level's AI Trainer Pay Index
AI Trainer Pay Index
The median advertised hourly rate for AI trainer, annotation and expert evaluation work was $68 per hour in October 2026 so far. These are rates written in job listings, not what any person earns.
Updated Oct 21, 2026
- Median advertised rate per hour
- $68
- Listings behind it
- 113
- Platforms with a published figure
- 1
micro1 provides most of the rates behind this month, so the overall figure mostly describes micro1. Only one month has enough rates, so there is no trend to read yet.
How do advertised rates change by month?
| Month | Platform | Listings | With a rate | Median per hour | Middle half | Basis |
|---|---|---|---|---|---|---|
| October 2026 so far | All platforms | 154 | 113 | $68 | $45 to $93 | Observed |
| October 2026 so far | micro1 | 143 | 112 | $68 | $45 to $93 | Observed |
| September 2026 | All platforms | 35 | 4 | Below 5 rates | Not shown | Observed |
| September 2026 | micro1 | 123 | 89 | $65 | $50 to $100 | Backfilled |
| August 2026 | micro1 | 26 | 18 | $66 | $53 to $79 | Backfilled |
| August 2026 | Other employers | 17 | 14 | $40 | $40 to $40 | Backfilled |
| April 2026 | Other employers | 17 | 17 | $40 | $40 to $40 | Backfilled |
The middle half runs from the first to the third quartile. A figure appears only from 5 listings with a rate.
Which expertise has the highest advertised rate?
Listings from every month Level has read, grouped by the expertise in the title. Domains with fewer than 5 rates show no figure.
micro1 supplies most of these rates, so the domain medians mostly describe micro1.
| Expertise | Listings | With a rate | Median per hour | Middle half | Lowest to highest |
|---|---|---|---|---|---|
| Law | 12 | 11 | $125 | $75 to $135 | $65 to $135 |
| Finance | 14 | 8 | $100 | $89 to $104 | $40 to $135 |
| Medicine | 29 | 21 | $98 | $68 to $108 | $26 to $125 |
| Software | 22 | 14 | $74 | $65 to $83 | $48 to $92 |
| Sciences | 12 | 8 | $68 | $68 to $90 | $45 to $100 |
| Business and operations | 20 | 6 | $48 | $34 to $52 | $25 to $55 |
| Generalist | 55 | 35 | $46 | $35 to $59 | $15 to $140 |
| Languages | 7 | 6 | $25 | $18 to $27 | $16 to $43 |
| AI and machine learning | 10 | 4 | Below 5 rates | Not shown | Not shown |
| Applied engineering | 8 | 4 | Below 5 rates | Not shown | Not shown |
Which platforms are missing from the figures?
- Mercor has 40 listings in Level's database, with 1 showing an hourly rate.
- Alignerr, DataAnnotation, Outlier: no listings in Level's database, so no figure. This says nothing about what they pay.
For what each platform publishes about its own pay, see AI training platforms compared. For open roles, see the AI trainer jobs hub.
Download the data
One row per month, platform and expertise domain, with the number of listings and every figure shown above. Free to reuse with attribution, under CC BY 4.0.
Common questions
Which expertise pays the most for AI training work?
Law has the highest median advertised rate among the domains that pass the minimum sample, at $125 per hour. That figure rests on 11 listings. Domains with fewer listings than the minimum are not shown, so the list will grow as more listings are read.
Is the index what an AI trainer will earn?
No. The index reads the hourly rate written in a job listing. Many platforms pay per task or limit the hours, so what a person earns depends on the work they are given.
Which platforms does the index cover?
It covers the listings Level can read from micro1, Mercor, Alignerr, DataAnnotation and Outlier, plus other employers that post AI training roles. A platform with no listings in Level's database has no figure. That says nothing about what it pays. See AI training platforms compared for what each platform publishes on its own pages.
How is the AI Trainer Pay Index calculated?
Level takes the hourly rate of each listing in its database, converts it to US dollars, and groups listings by month, platform and expertise. A group shows its median and quartiles only from 5 listings. The method block below the chart gives the sources, the period, the conversion and the limits.
Method
- Scope
- Live and expired job listings in Level's database for AI trainer, annotation and expert evaluation work: the data annotation category, listings read from micro1, and listings from Mercor, Alignerr, DataAnnotation and Outlier. Other employers in the category are grouped as Other employers.
- Sample
- 480 listings, of which 269 advertise an hourly rate that passes the checks. Level read the listings from company career pages and platform feeds. Nothing was collected from a third-party site that forbids it.
- Measured
- Read from Level's database each time the page is built, refreshed every 10 minutes, on the date shown under the title. A listing counts in the month it was posted, in UTC. Level's database starts on 2026-09-09 and first read these sources in September 2026.
- Currency
- US dollars per hour. A rate in another currency is converted through euros, at 1.16 US dollars to the euro, the rounded table Level uses for its salary checks. A rate with no currency is left out.
- Not measured
- What anyone earns. Advertised rates are not earnings, many platforms pay per task, and a listing can describe a rate for a narrow project. A listing that Level did not read is not counted.
- Rate
- The middle of the range written in the listing, read as an hourly rate only when its size says so. Amounts that could be monthly or annual are left out. Rates below $10 per hour are left out as typing mistakes or per-task prices.
- Expertise
- Each listing is placed in one domain from its title, then from the start of its description: law, software, medicine, languages, finance, sciences, applied engineering, AI and machine learning, business and operations, or generalist when nothing more specific matches. This is a keyword rule, so some listings are placed imperfectly.
- Minimum sample
- A group shows its median, quartiles, lowest and highest rate only from 5 listings with a rate. Below that, the file keeps the counts and leaves the figures empty.
- Backfilled months
- A month before Level first read a platform is marked backfilled. Those listings were still open when Level arrived, so short-lived listings are missing. Backfilled months are shown apart, and they never enter the all-platform figures.
- Limits
- One platform can supply most of a month, which the page says when it happens. A platform with no listings in Level's database has no figure, which says nothing about what it pays.
Cite this report
Level's AI Trainer Pay Index can be quoted directly. Please attribute it as a report name, not as the Level brand.
Cite this page
Level's AI Trainer Pay Index. Data measured on Oct 21, 2026. https://jobsbylevel.com/ai-trainer-pay-index