Every AI training platform will tell you what its best-paid people earn. Almost none will tell you what the typical open role actually advertises. So we measured it. On October 1, 2026 we read every active listing in our feed, 3,821 roles across 19 platforms, and pulled out the ones that publish an hourly rate in US dollars: 2,537 listings.
The headline is that the median listing advertises a top rate of $65/hr, and the middle half of listings top out between $27 and $100 an hour. That is a very wide band, and most of this report is about what sits behind it: the platform, the profession, the language, the country, and the credentials a listing asks for.
Three things to keep in mind throughout. These are advertised rates, meaning what a listing says it pays, not what anyone took home. This is the first edition, so there is no trend yet, only a baseline to measure against. And the live version of these numbers on our pay report page updates every day, so it will drift away from the fixed snapshot you are reading.
The short version
- The median is $65/hr, but the shape matters more than the middle. 61% of listings advertise at least $50 an hour and 25% at least $100, while 9% advertise under $20.
- Platform shows the widest spread. The highest platform median (Babel Audio, $110/hr) is 11.0× the lowest (Welo Data, $10/hr).
- Four professions tie at $100 an hour: engineering, finance and accounting, law and cybersecurity. Languages and translation, by far the largest category, sits at $30/hr.
- Credentials pay. Listings that require a professional licence advertise a median of $104/hr against $65/hr for those that do not. Listings that require a doctorate advertise $92/hr, and those that require coding $90/hr.
- Entry-level work pays well under half the market rate.Beginner-friendly listings advertise a median of $26/hr, about 37% of the $70/hr median for everything else.
- Where you are changes what is offered. Worldwide listings advertise a median of $65/hr; listings limited to named countries advertise $45/hr.
- The ceiling is not the offer. Where a listing states a minimum, the median minimum is $35/hr, roughly half the median top rate.
- Flat-fee studies are a separate market:112 listings with a median of $31 per project, which cannot be compared with an hourly rate.
How we measured pay, and what we left out
The figures come from the same method as the live pay report, which is described in full on that page. In brief:
- Only USD hourly rates count. A listing has to state an hourly rate in dollars to be included in a median. 2,537 of the 3,821 active listings qualify.
- We use the top of the range. A listing that says “$40 to $80 an hour” counts as $80. Every priced listing states a top figure, which makes platforms comparable. Where a minimum is stated we report it separately.
- Flat fees are kept apart. 112 listings are priced per project, mostly paid studies and interviews. A $400 study that takes an hour is not the same rate as a $400 project that takes a week.
- Not counted: 930 listings that state no pay at all, and 242 priced in other currencies, per year, or outside the range we treat as a real hourly wage. We also set aside one listing that advertised $1,000 an hour for about twenty hours a week of financial modelling, which contradicted its own description, so we treated it as a probable typo at the source.
- Small groups are hidden. A platform or profession needs at least five priced listings before we publish a figure for it.
- Low rates stay in. The lowest rate in the data is $2.40 an hour. Some platforms advertise a few dollars an hour for regional language and ad-rating tasks. Hiding them would overstate what the market pays.
Why we balance by platform
Listing counts are wildly uneven, and that can distort a median.Meridial alone accounts for 826 of the 2,537 priced listings, which is 33% of the whole, and 654 of those share exactly the same $65 ceiling. Counted listing by listing, one platform's template would set the market median by itself.
So every overall, per-profession, per-language and per-country figure in this report is platform-balanced: no single platform may carry more than 15% of the weight in an aggregate. The per-platform table is not weighted at all, because each row is that platform on its own.
How much does it change? Less than you might expect at the centre and more at the edges. The overall median is $65/hr without balancing and $65/hr with it. Drop Meridial entirely and it is $60/hr; take the median of the twelve platform medians and it is $70/hr. So the headline is stable: somewhere between $60 and $70 however you slice it. What balancing changes is the spread. The top quartile boundary moves from $80 unbalanced to $100 balanced, and some professions shift a lot: customer support goes from $6 to $55, cybersecurity from $62 to $100, while law moves the other way, from $116 to $100.
The headline numbers
Across 2,537 priced listings, the median top rate is $65/hr. The lowest quarter of listings top out at $27 or less; the highest quarter at $100 or more. The single highest rate in the data is $400/hr.
The distribution is more informative than the median. Here is how priced listings spread across rate bands, both counted listing by listing and platform-balanced:
| Top rate | Listings | Share (listing by listing) | Share (platform-balanced) |
|---|---|---|---|
| Under $15 | 126 | 5% | 5% |
| $15–$29 | 493 | 19% | 21% |
| $30–$49 | 293 | 12% | 14% |
| $50–$74 | 892 | 35% | 24% |
| $75–$99 | 220 | 9% | 11% |
| $100–$149 | 341 | 13% | 17% |
| $150–$199 | 95 | 4% | 5% |
| $200 and over | 77 | 3% | 3% |
The difference between the two columns is almost entirely the $50 to $75 band. Listing by listing it holds 35% of the market, because a single templated $65 ceiling lands there hundreds of times. Balanced, it holds 24%. Either way, the market is not a bell curve around $65. It is a sizeable low end (about 25% of listings under $30), a broad middle, and a thin but real expert tail: 77 listings advertise $200 an hour or more.
That expert tail is concentrated. Of those 77, 53 come from micro1 and 20 from Mercor. By profession, 18 are medicine, 17 law, 13 science and research, and 8 finance and accounting. If you want to be paid $200 an hour to train AI, the market is telling you which doors to knock on.
Pay by platform: a spread of 11×
Here are the twelve platforms with at least five priced listings, ranked by median top rate. “Middle half” is the 25th to 75th percentile of top rates. “Median minimum” appears where at least five listings state one.
| Platform | Priced listings | Median top rate | Middle half | Median minimum | Highest |
|---|---|---|---|---|---|
| Babel Audio | 43 | $110/hr | $110–$150 | $60/hr | $150/hr |
| micro1 | 336 | $91/hr | $60–$140 | $50/hr | $400/hr |
| Mercor | 314 | $90/hr | $55–$119 | $60/hr | $250/hr |
| Braintrust | 23 | $85/hr | $26–$95 | $50/hr | $150/hr |
| AfterQuery | 209 | $75/hr | $45–$103 | $75/hr | $175/hr |
| Terac | 17 | $75/hr | $28–$100 | $75/hr | $260/hr |
| Meridial | 826 | $65/hr | $65–$65 | $8/hr | $200/hr |
| SME Careers | 260 | $50/hr | $25–$70 | — | $150/hr |
| xAI | 33 | $45/hr | $45–$45 | $35/hr | $100/hr |
| Prolific | 389 | $26/hr | $26–$40 | $20/hr | $120/hr |
| Turing | 20 | $15/hr | $15–$25 | $15/hr | $200/hr |
| Welo Data | 61 | $10/hr | $7–$20 | $10/hr | $55/hr |
The premium tier: $75 and up
Babel Audio has the highest median ($110/hr) with a notably tight band of $110–$150: voice and audio specialists, priced consistently. micro1($91/hr) and Mercor ($90/hr) are the two big expert platforms and the volume story of the premium tier, with 336 and 314 priced listings. They differ in shape. Mercor's middle half is $55–$119 with a median minimum of $60. micro1 reaches higher ($400/hr at the top, the highest in the data) but is also wider at the bottom, with a middle half of $60–$140.
Braintrust ($85/hr, only 23 priced listings), AfterQuery($75/hr, 209 listings) and Terac ($75/hr, 17 listings) complete the tier. Treat Braintrust and Terac with care: with fewer than thirty priced listings each, one new batch of postings can move the median noticeably. AfterQuery is the cleanest of the three, with a median minimum of $75 equal to its median top rate, which suggests many of its listings state a single fixed rate and not a range.
The middle: $45 to $65
Meridial sits at $65/hr and deserves a closer look, because the headline hides the real shape. Its median minimum is $8: the platform advertises a band from roughly $6 up to $65, and the $65 end is what we count as the top rate. Our Meridial profile notes that the exact rate is set after weighing experience and location, so the ceiling is best read as a maximum rather than a typical offer. SME Careers($50/hr, 260 priced listings) almost never states a minimum, so there is no floor to report.xAI ($45/hr) is a narrow, consistent band.
The low end: under $30
Prolific ($26/hr, 389 priced listings) is the second-largest source of priced listings and sits well below the market median: participant-recruitment work, language and developer tasks, priced by the hour.Turing shows $15/hr, but be careful: only 20 of its 116 active listings state a rate, so that median describes only those: at least three quarters top out at $25 or less, with a few specialist roles reaching as high as $200/hr. It says little about what Turing pays overall. Welo Data($10/hr, 61 listings, top at $55/hr) is the lowest of the twelve: multilingual annotation and evaluation, including regional ad-rating roles, with a quarter of its priced listings topping out at $7 or less.
Volume is not pay
The two platforms with the most priced listings, Meridial (826) and Prolific (389), account for 48% of all priced listings, and their medians are $65/hr and $26/hr. The size of a job board is a poor guide to what it pays: the third and fourth largest by priced listings, micro1 andMercor, are the two big premium-tier platforms, while the two largest sit at the middle and low end of the market.
Pay by profession: a four-way tie at the top
Every listing in our feed is classified by the background it asks for. Here are the eighteen categories with at least five priced listings, platform-balanced. Each name links to the matching live roles.
| Profession | Priced listings | Median top rate | Middle half | Highest |
|---|---|---|---|---|
| Engineering | 182 | $100/hr | $65–$120 | $175/hr |
| Finance & accounting | 114 | $100/hr | $68–$112 | $350/hr |
| Law | 102 | $100/hr | $80–$129 | $400/hr |
| Cybersecurity | 40 | $100/hr | $68–$101 | $250/hr |
| Medicine & healthcare | 116 | $90/hr | $65–$150 | $285/hr |
| Science & research | 192 | $80/hr | $65–$104 | $250/hr |
| Design & creative | 165 | $75/hr | $45–$100 | $250/hr |
| Mathematics | 37 | $75/hr | $65–$88 | $110/hr |
| Software engineering | 270 | $70/hr | $45–$100 | $350/hr |
| Business & consulting | 76 | $70/hr | $50–$100 | $350/hr |
| Data science | 64 | $70/hr | $39–$120 | $350/hr |
| Humanities | 15 | $66/hr | $65–$75 | $75/hr |
| Education | 23 | $65/hr | $51–$77 | $140/hr |
| Writing & editing | 22 | $65/hr | $37–$72 | $140/hr |
| Customer support & admin | 41 | $55/hr | $23–$69 | $92/hr |
| Miscellaneous | 78 | $50/hr | $20–$70 | $200/hr |
| Languages & translation | 971 | $30/hr | $24–$55 | $110/hr |
| General (no specialty) | 29 | $20/hr | $18–$52 | $200/hr |
Four categories tie at a $100 median: engineering (excluding software), finance and accounting, law and cybersecurity. The tie hides real differences in shape. Law has the highest floor and the highest ceiling: its bottom quartile starts at $80 (the others at $65 to $68) and its top single rate is $400/hr, the highest of any profession. Medicine's median is $90/hr, slightly lower, but its top quartile runs furthest of all, to $150. Cybersecurity is the smallest of the four with 40 priced listings.
Then the biggest category by far: languages and translation, with 971 priced listings. That is 38% of everything priced, at a median of $30/hr. It is the main on-ramp for multilingual people, and it is also why the overall distribution has such a sizeable low end. Generalist roles with no particular specialty sit at $20/hr on 29 priced listings. Software engineering is large (270 priced listings) and lands at $70/hr, below the top four, because it spans a huge range by platform, from $100/hr on micro1down to $15/hr onTuring (see the next section).
The same job pays differently on different platforms
Profession is not the whole story. Take one profession and look at what each platform advertises for it (platforms with at least five priced listings in that profession):
| Profession | micro1 | Mercor | AfterQuery | SME Careers | Meridial | Other platforms |
|---|---|---|---|---|---|---|
| Law | $135/hr (40) | $120/hr (16) | $121/hr (14) | $113/hr (8) | $68/hr (20) | — |
| Finance & accounting | $110/hr (19) | $110/hr (25) | $110/hr (18) | $85/hr (7) | $8/hr (39) | — |
| Medicine & healthcare | $200/hr (25) | $119/hr (23) | $85/hr (26) | $90/hr (12) | $65/hr (28) | — |
| Software engineering | $100/hr (36) | $90/hr (37) | $80/hr (15) | $65/hr (51) | $65/hr (105) | Terac $75/hr (5), Turing $15/hr (15) |
| Science & research | $200/hr (19) | $105/hr (38) | $90/hr (13) | $78/hr (20) | $65/hr (43) | Prolific $40/hr (58) |
The platform effect is large. For law, micro1's median ($135/hr) is about 2.0× Meridial's ($68/hr). For medicine, it is 3.1× ($200/hr against $65/hr). In finance, micro1, Mercor and AfterQuery all advertise the same $110/hr median.
One entry needs a warning. Meridial's finance median looks like $8/hr, which would be absurd as a pure platform comparison. It is not an error; it is geography. Many of Meridial's finance roles, such as custody operations and retirement plan administration, are country-specific and priced at about $5 an hour for countries like Colombia, Jamaica and the Philippines, while its worldwide roles go up to $65. Meridial prices by location, so the same job title can carry a very different rate depending on the country named. Read its row as “this platform prices by place” rather than “this platform pays finance people $8.”
What actually moves the rate: credentials, code and language
Beyond profession, each listing is tagged for requirements. Comparing listings that carry a requirement with those that do not:
| Listing requirement | Median with it | Listings | Median without | Listings | Ratio |
|---|---|---|---|---|---|
| Requires a professional licence | $104/hr | 123 | $65/hr | 2,414 | 1.61× |
| Requires a doctorate | $92/hr | 163 | $65/hr | 2,374 | 1.41× |
| Requires coding | $90/hr | 329 | $62/hr | 2,208 | 1.45× |
| Open to beginners | $26/hr | 568 | $70/hr | 1,969 | 0.37× |
| Needs a non-English language | $33/hr | 1,181 | $75/hr | 1,356 | 0.44× |
| Needs voice or camera | $30/hr | 200 | $65/hr | 2,337 | 0.46× |
A professional licence is worth the most: listings that require one advertise 1.6× the median of those that do not. A doctorate requirement shows a 41% premium, and coding a 45% premium. On the other side, beginner-friendly listings advertise $26/hr, and listings that need a voice or camera advertise $30/hr.
A caution on reading this. These are associations, not causes. Listings that require a licence also tend to sit on premium platforms and in premium professions, so the licence is not the only thing doing the work. But the direction is consistent with everything else in the report: the market pays for verified, scarce expertise, and pays far less for work that anyone can start tomorrow.
What kind of work
The task itself matters too. Among task types with at least twenty priced listings:
| Type of work | Priced listings | Median top rate | Middle half |
|---|---|---|---|
| Transcription | 38 | $93/hr | $30–$110 |
| Expert consultation | 158 | $86/hr | $61–$100 |
| Writing training content | 160 | $76/hr | $35–$105 |
| Software demos | 61 | $72/hr | $40–$120 |
| Red-teaming | 418 | $65/hr | $54–$92 |
| Voice recording | 145 | $45/hr | $30–$100 |
| Rating AI answers | 938 | $40/hr | $20–$85 |
| Data annotation | 65 | $20/hr | $10–$45 |
“Rating AI answers” is the entry route and by far the largest bucket (938 priced listings), at a median of $40/hr. Expert consultation ($86/hr) and writing training content ($76/hr) sit well above it. Data annotation, the most routine work, is the lowest at $20/hr. The transcription row is a small sample (38 listings), so we would not read much into it.
Languages: plenty of work, lower advertised rates
Listings that require a language other than English advertise a median of $30/hr (1,116 priced listings), against $75/hr for listings with no language requirement (1,421 priced listings). Some of that is composition, since listings that require a language are not spread evenly across platforms. But it means language alone is a weak lever.
| Language | Listings | Priced listings | Median top rate | Middle half |
|---|---|---|---|---|
| Swedish | 33 | 22 | $52/hr | $30–$62 |
| German | 76 | 50 | $50/hr | $29–$65 |
| Dutch | 47 | 23 | $45/hr | $23–$65 |
| Japanese | 55 | 34 | $40/hr | $30–$65 |
| Korean | 65 | 29 | $40/hr | $29–$65 |
| Spanish | 111 | 59 | $40/hr | $20–$65 |
| French | 134 | 71 | $35/hr | $25–$65 |
| Portuguese | 53 | 31 | $35/hr | $20–$65 |
| Arabic | 49 | 29 | $34/hr | $15–$65 |
| Mandarin | 41 | 38 | $30/hr | $18–$65 |
| Marathi | 39 | 36 | $35/hr | $25–$117 |
| Tamil | 55 | 52 | $25/hr | $20–$65 |
| Hindi | 37 | 30 | $24/hr | $20–$65 |
Within languages the pattern is clear. Swedish ($52/hr), German ($50/hr) and Dutch ($45/hr) lead. The large Western European languages sit between French at $35/hr and Spanish at $40/hr. Three South Asian languages are at the bottom: Hindi at $24/hr, Tamil at $25/hr and Punjabi at $26/hr. Marathi is a curiosity: a median of $35/hr but a middle half running from $25 to $117, the widest of any language, so some Marathi roles pay like specialist work.
The practical advice is to treat language as a qualifier, not a profession. A fluent French speaker who is also a lawyer or a software engineer is in a different market from one who is applying on language alone. You can browse every language on our browse page and combine it with a profession filter.
Geography: where you are changes what is offered
Worldwide listings, which state no country restriction, advertise a median of $65/hr (1,451 priced listings). Listings that name specific countries advertise $45/hr(1,086 priced listings). Here are the countries with the most priced listings that name them:
| Country | Listings naming it | Priced listings | Median top rate | Middle half |
|---|---|---|---|---|
| United States | 492 | 384 | $65/hr | $40–$90 |
| United Kingdom | 69 | 59 | $65/hr | $20–$97 |
| Canada | 72 | 55 | $61/hr | $20–$82 |
| Germany | 103 | 85 | $65/hr | $30–$90 |
| Singapore | 81 | 77 | $65/hr | $20–$65 |
| Brazil | 117 | 68 | $65/hr | $7–$65 |
| Indonesia | 87 | 77 | $65/hr | $9–$65 |
| Philippines | 101 | 90 | $30/hr | $6–$65 |
| Malaysia | 89 | 80 | $20/hr | $7–$60 |
| Pakistan | 99 | 98 | $62/hr | $20–$65 |
| Bangladesh | 95 | 94 | $55/hr | $15–$65 |
| Thailand | 80 | 77 | $15/hr | $8–$65 |
| Vietnam | 68 | 66 | $16/hr | $5–$65 |
| India | 236 | 192 | $11/hr | $7–$65 |
Read this table carefully, because two things are happening at once. First, the median for 6 of the 14 countries in this table is exactly $65. That is Meridial's templated ceiling again, repeated across many countries. The telling column is the middle half. For Brazil it runs from $7 to $65 and for Indonesia from $9 to $65: the advertised ceiling is the same, but a large share of the roles sit far below it.
Second, some countries really are different. Roles that name India have a median of $11/hr, and Thailand, Vietnam and Malaysia are between $15/hr and $20/hr. India is the starkest case: 192 priced listings, of which 125 come from SME Careers and 52 fromMeridial. The United States, by comparison, has 384 priced listings with a median of $65/hr and a middle half of $40–$90.
The takeaway is not that your country determines your pay. It is that the roles that name your country are a different pool from the worldwide pool, and you should look at both. Each country has its own page with live listings, such as the United States or India, and the full list is on the countries page.
Flat-fee studies: a different market
112 listings are priced per project rather than per hour: paid research studies, interviews and short tasks. Their median fee is $31 and the middle half runs from $19 to $126. 58 of them pay under $50 and 16 pay $200 or more, with the largest at $3,000. They come mainly from Terac (45), Prolific (35), Braintrust(15) and Mercor (11).
The only fair way to compare a flat fee with an hourly role is to divide by the time it takes. A $400 interview that runs an hour is a $400 rate; a $400 project that takes a week is not. Always check the listed duration before you decide a study is “high paying,” and equally before you dismiss a low fee that takes ten minutes.
The extremes: $2.40 and $400
The lowest advertised hourly rate in the data is $2.40; the highest is $400. At the bottom, 98 listings advertise under $10, and they come almost entirely from two sources: Meridial (64) and Welo Data (26), both of which price regional language and rating work by country. At the top, 77 advertise $200 or more, almost all from micro1 and Mercor, and mostly for physicians, lawyers, scientists and finance professionals.
The gap between those two numbers, a factor of 167×, is the clearest picture of how different the corners of this market are. “AI training job” describes both a regional ad-rating task and a specialist physician review, and they have almost nothing in common except the label.
Why these numbers can mislead
- Advertised is not earned. A listing's rate says nothing about how many hours the platform will give you. Work arrives in projects that start and stop, so a high rate with sporadic hours can earn less than a modest rate with steady ones.
- Ceilings flatter. The median top rate is $65/hr, but among listings that state a minimum the median minimum is $35/hr. For the 2,279 listings that state a minimum, the median ratio between top rate and minimum is only 1.5×, but a quarter of them have a top rate more than 8.1× their minimum. A wide band means the headline shows the best case, not what you will necessarily be offered.
- Prices vary by place. As Meridial shows, the same job can carry different rates by country. A single platform median can hide that.
- Classification is automatic. Professions, languages and requirements are assigned by reading each listing's text. It is accurate in bulk and occasionally wrong on individual listings, and the smaller categories (a dozen or so listings) are most exposed to a mistake or two.
- A listing is not a hire. One role posted for twenty countries counts as twenty listings. Balancing by platform reduces, but does not remove, that effect.
- It only covers platforms we track. Our feed has 19 platforms with active listings. Work arranged outside them, or through employers who do not post publicly, is invisible here.
What this means for you
If you have professional credentials
You are in the part of the market the data rewards most. Law, medicine, finance and engineering all advertise a $90 to $100 median, and the best-paid listings reach $200 to $400. Compare platforms before you apply: in law and medicine the platform medians differ by a factor of two or three. Start with the legal, medical, finance and software pages, and read the screening notes in our interview guide.
If you are starting out
Be realistic. Beginner-friendly listings advertise a median of $26/hr, and rating AI answers sits at $40/hr. That is a legitimate way to build a track record, but it is not the $100-plus end of the market. Our beginner's guide and the no experience page are the places to start, and our scam guide is worth reading first: no legitimate platform charges you to apply.
If you are multilingual or outside the US
Pair your language with something else. A language on its own advertises $30/hr; combined with a profession it can move you into a different pool. Check both the roles that name your country and the worldwide roles, and read each listing's eligibility before you apply. See the country pages and our guide to working from outside the US.
A quick way to use this report
- Find your profession in the table and note its median and middle half.
- Look at which platforms advertise the most for it, and how many listings they have.
- Check the minimum, not just the top rate, and treat a wide band as a negotiation.
- Open the live roles on the online AI training jobs page and compare the actual listings with the medians here.
What we will track next
This is the first snapshot, so everything here is a level, not a trend. The live pay report refreshes daily, and once there is a second snapshot to compare against, the next edition will show month-over-month change by platform and profession: which platforms are raising or cutting advertised rates, whether the expert tail is growing, and whether more listings start publishing a rate at all. That last one matters more than it sounds. 930 of our 3,821 listings, about 24%, state no pay, and the more that change, the more reliable every number in this report becomes.
Common questions
What is the average hourly rate for AI training jobs?
The median advertised top rate is $65/hr across 2,537 listings, with the middle half between $27 and $100. The range is wide enough that an average is not very useful on its own. Your profession and the platform matter far more.
Which AI training platform pays the most?
By median top rate among platforms with at least five priced listings,Babel Audio is highest at $110/hr, followed bymicro1 ($91/hr) and Mercor($90/hr). By highest single advertised rate, micro1 leads at $400/hr. Rankings change as listings come and go; the live pay report is always current.
Do AI training jobs pay by the hour?
Most do: 2,537 of our listings state an hourly rate. A further 112 are flat fees per study or project, and 930 state no pay at all, with pay agreed after you apply.
Do you need a degree to earn more?
Credentials clearly pay: licence-required listings advertise $104/hr and doctorate-required ones $92/hr. But skills count too. Listings that require coding advertise $90/hr without needing a particular degree.
How accurate are these figures?
They accurately describe what listings advertise on the day we measured, using a method we publish. They do not describe earnings, and the smallest groups are sensitive to a handful of listings. We would rather show you the sample size next to every figure than claim precision we do not have.
Data as of October 1, 2026. Method and live figures: AI training pay report.