AI training pay dataset
Short answer: of 992 live AI training listings on 1 November 2026, 660 state an hourly rate. The median low end is $60 and the median high end is $92 an hour. These are advertised rates from Mercor, Turing, micro1, not earnings. The listing-level data is free to download below.
Data as of 1 November 2026. Rebuilt with every site update, so the numbers move as listings open and close. If you quote a figure, quote the date with it.
Download
- ai-training-pay-jobs.csv one row per live listing (992 rows)
- ai-training-pay-jobs.json the same rows as JSON, with the date and licence
How to cite
Licensed CC BY 4.0. Please credit the source, give the date, and link here.
AI Work Expert (2026). AI training job pay dataset, data as of 1 November 2026. https://www.aiworkexpert.com/data/ai-training-pay. Licensed CC BY 4.0.
Key figures
| Pay unit | Listings | Median low | Median high | Lowest to highest |
|---|---|---|---|---|
| Per hour | 660 | $60/hr | $92/hr | $6 to $1,000 |
| Per year | 41 | $280,000/yr | $450,000/yr | $15,000 to $6,000,000 |
| Per task or one-time | 37 | $65 per task | $75 per task | $60 to $2,200 |
254 of 992 live listings state no pay and are left out of every figure. The lowest and highest figures are single listings as advertised, so read the medians as the typical case.
Hourly pay by platform
| Platform | Listings | Median low ($/hr) | Median high ($/hr) |
|---|---|---|---|
| Mercor | 314 | $60 | $100 |
| micro1 | 325 | $50 | $90 |
| Turing | 21 | $15 | $15 |
Groups with fewer than 5 listings are not shown. Yearly and per-task pay is reported separately: see micro1 (38 yearly listings, median $300,000 to $500,000); and Mercor (37 per-task listings, median $65 to $75 per task).
Hourly pay by category
| Category | Listings | Median low ($/hr) | Median high ($/hr) |
|---|---|---|---|
| Research | 70 | $70 | $110 |
| Coding | 66 | $60 | $103 |
| Writing & feedback | 27 | $50 | $100 |
| Model evaluation | 444 | $60 | $90 |
| Data annotation | 53 | $25 | $50 |
Hourly pay by industry
| Industry | Listings | Median low ($/hr) | Median high ($/hr) |
|---|---|---|---|
| Legal | 78 | $90 | $145 |
| Healthcare | 68 | $78 | $118 |
| Research | 43 | $70 | $110 |
| Finance | 60 | $70 | $105 |
| Developer tools | 62 | $60 | $100 |
| Consulting | 8 | $63 | $95 |
| Government | 22 | $40 | $80 |
| Media | 56 | $50 | $80 |
| Marketing | 17 | $50 | $70 |
| AI / ML | 246 | $48 | $65 |
Category, role and industry are our own labels, assigned by keyword rules, so they are less exact than the pay figures.
How we built it
- We read each live listing on Mercor, micro1 and Turing and record its stated pay, type, location and dates.
- A listing is live until its expiry date. Closed listings are removed, so the dataset is a snapshot of what is open on the date shown.
- A range such as $20 to $30 an hour gives a low and a high. A single figure gives the same number for both.
- Only pay a listing states is used. A listing that says "Pay varies" or states nothing is counted as having no stated pay. Nothing is estimated or filled in.
- Hourly, yearly and per-task pay stay in their own units and are never converted.
- Each listing counts once. Medians are calculated across listings, so a platform with many similar listings weighs more than one with a few.
- Groups with fewer than 5 listings are not published.
What this data is not
- Advertised pay is not earnings. Workers may not get steady hours, and pay can change after hire.
- It covers three platforms. It is not a survey of the whole AI training market.
- Platforms may edit or remove a listing after we read it.
- Job titles and pay figures are reproduced as facts. The platforms keep their rights in their own listing text, which this dataset does not include.
Columns
| job_id | Our stable id for the listing. The page is at /jobs/{job_id}. |
| platform | The platform that posted the listing (Mercor, micro1 or Turing). |
| title | The listing title. |
| category, role, industry | Our own classification of the listing, assigned by keyword rules. Treat it as a guide, not a platform field. |
| work_type | Hourly, Task-based or Project, as shown on the listing. |
| pay_stated | yes if the listing states a pay figure, no if it does not (for example "Pay varies"). |
| pay_unit | hour, year or task. A task rate also covers one-time flat payments. Units are never converted into each other. |
| pay_min_usd, pay_max_usd | The low and high end of the stated range in US dollars. A single stated figure appears as both. |
| location, contract_type | As stated on the listing. |
| posted_at, expires_at | ISO 8601 timestamps. expires_at is when we treat the listing as closed. |
| page_url | The listing's page on this site. |
Questions about this data
How much do AI training jobs pay per hour?
In 1 November 2026's data, 660 live listings state an hourly rate. The median low end is $60 and the median high end is $92 an hour. These are advertised rates, not what workers earn, and the range runs from $6 to $1,000 across all listings.
Where does the data come from?
From the listings tracked on Mercor, Turing, micro1, each read from the platform's own page. Only pay that a listing states is used. Nothing is estimated.
Can I use this data?
Yes. The files are free to copy, share and adapt under the Creative Commons Attribution 4.0 licence (CC BY 4.0). Credit AI Work Expert, quote the date, and link to this page.
Why are per-task and yearly pay not mixed with hourly pay?
A task payment, a one-time flat payment and an annual salary are different units. Converting them would require assumptions about hours worked, so the dataset reports each unit on its own.
Related
- Mercor pay rates
- micro1 pay rates
- How much do AI training jobs pay?
- Browse live AI training jobs
- About AI Work Expert and how we collect data
AI Work Expert is an independent publisher. Some outbound links to platforms are referral links, which does not change the figures above.