Pay
What your specialty is actually worth in AI training
In this line of work, what you get paid tracks the scarcity of your expertise far more than the hours you put in. That claim gets made a lot and almost never shown. It can be shown here, because one platform — micro1 — publishes pay on nearly every role on its own public job board, so the rate gap between a lawyer, an engineer, and a generalist is a matter of arithmetic rather than assertion.
What the board actually pays, by specialty
Computed from the 295 of 317 open roles that publish an hourly rate, mirrored from micro1's own board on October 28, 2026. Board-wide, the median listing sits at $68/hr at the midpoint of its range — half the priced roles pay more, half less — and the middle half of listings falls between $47 and $108 an hour. The lowest figure any listing publishes is $6 and the highest is $400: those are the two ends of the board, not a range to plan around. Rows are ordered richest first, by each specialty's median midpoint — the same figure the table leads with.
| Specialty | Priced listings | Median /hr | Middle half /hr | Full spread /hr |
|---|---|---|---|---|
| Sciences & Research | 23 | $150 | $58–$150 | $7–$400 |
| Law | 51 | $130 | $95–$270 | $40–$400 |
| Medicine | 21 | $98 | $60–$103 | $20–$125 |
| Finance | 18 | $90 | $90–$238 | $30–$350 |
| AI & Machine Learning | 12 | $90 | $44–$111 | $7–$300 |
| Software Engineering | 26 | $73 | $61–$75 | $7–$300 |
| Applied Science & Engineering | 14 | $68 | $60–$68 | $45–$150 |
| Business Operations | 14 | $66 | $56–$68 | $15–$350 |
| Data Analysis | 6 | $63 | $48–$78 | $25–$350 |
| Other roles | 48 | $60 | $46–$74 | $7–$140 |
| Generalist | 19 | $28 | $24–$40 | $6–$140 |
| Language & Audio | 16 | $27 | $16–$46 | $8–$98 |
| Sales & Marketing | 5 | $25 | $18–$30 | $7–$350 |
| Robotics | 13 | $20 | $9–$23 | $8–$120 |
“Median” reduces each listing to the midpoint of its own range and takes the middle one, so half a specialty's priced roles sit above it and half below. “Middle half” is the band the central 50% of those midpoints fall in: a quarter of the specialty's listings are priced below it and a quarter above, which makes it the closest thing on this page to a typical range. The full spread is the two extremes, and most listings sit nowhere near either end of it. Every figure is pay micro1's own board advertises for a role, not pay anyone has been shown to receive. How a listed range converts into real, sustained take-home is a separate question, and nobody publishes an answer to it; the micro1 review sets out what micro1 does and does not commit to between a listed rate and a paid task.
The figures above leave out 5 of the board's specialties on purpose, for having too few priced listings to generalize from: a median drawn from a couple of roles is an anecdote wearing a statistic's clothes. Excluded, with their listing counts: Arts & Design (3), Retail & Hospitality (2), Education (2), Cybersecurity (1) and Humanities (1). If your field is one of those, roles still exist — go read the listings themselves rather than trusting an average.
Read the spread, not the median
The medians are the part everyone quotes, and they do say something real: Sciences & Research tops the board at $150/hr, the median of its 23 priced listings, against a board-wide median of $68/hr across all 295 priced roles on the board as mirrored on October 28, 2026. That is scarcity pricing, visible in published numbers.
Now look at the last column. Inside Sciences & Research alone, listings run from as little as $7 to as much as $400 an hour — a $393 gap between roles wearing the same specialty label, which is wider than the $130 that separates the board's highest and lowest specialty medians.
The practical reading: role-by-role variance dwarfs the difference between specialty averages. Your specialty tells you which door to knock on. It does not tell you what you will personally earn — that comes down to the individual listing, and two roles wearing the same specialty label can be priced nothing like each other. So use the table to decide where to spend your applications, then judge each role on its own posted number.
Where the money is, specialty by specialty
Working down the table from the top, here is what the strongest-paying specialties tend to involve and what tends to make a person credible in each. The work descriptions are general — how expert evaluation work is usually structured, not a description of any one platform's tasks — and every figure beside a heading comes from the same October 28, 2026 board as the table above. Specialties further down the table are no less real; they are simply priced lower on this board.
Sciences & Research
Grading reasoning rather than prose: walking a derivation step by step, catching a unit or sign error, checking that a cited result says what the model claims it says, writing worked solutions at graduate-problem-set level, and building adversarial problems where the plausible-looking answer is the wrong one.
What tends to count as evidence: Postgraduate training in a named field, publications or preprints, and the specific methods and instruments you've used. Subfield precision is the whole game — "single-cell RNA-seq in wet-lab molecular biology" is legible; "science background" isn't.
Law
Almost all of it is judgment applied to text: reading a model's answer about a contract, a statute, or a procedural question and saying exactly where it goes wrong; writing the reference answer that other answers get scored against; turning your own review instincts into a rubric someone less senior can apply; and hunting the failure modes only a practitioner catches — a confident citation to a case that doesn't exist, advice that ignores which jurisdiction it's in, a clause summary that quietly flips who carries the risk.
What tends to count as evidence: Your practice area does more work than your seniority. Name the qualification, the jurisdictions you're admitted in, and the documents you have actually drafted or litigated. Published writing and redacted work product you own the rights to are stronger than any adjective; "legal professional" with no specifics reads as generalist.
Medicine
Clinical judgment applied to text: rating whether an answer is safe as well as accurate, catching guideline drift and confidently wrong triage or dosing reasoning, writing reference answers with the clinical caveats left intact, and red-teaming the advice a patient might actually act on.
What tends to count as evidence: An active license and your specialty, stated plainly. Nurses, pharmacists, PAs, and allied clinicians are domain experts here, not near-misses — say which setting you practice in.
Finance
Checking numbers and the reasoning wrapped around them: whether a valuation walk holds together, whether an accounting treatment is defensible, whether a regulatory framing is current — and then writing the corrected version with the logic shown, not just the fixed figure.
What tends to count as evidence: Licensure and charters (CPA, CFA, actuarial credentials) are the most legible signals, followed by a modeling portfolio you're allowed to show and the specific instruments, markets, or reporting regimes you worked with.
AI & Machine Learning
Evaluating model behavior with a practitioner's eye: comparing candidate outputs against each other, writing eval sets and scoring guidelines other people can run, diagnosing where a chain of reasoning actually breaks, and probing for failure modes reproducibly instead of anecdotally.
What tends to count as evidence: Shipped work beats vocabulary. Repositories, eval or model artifacts, papers, and production systems you were responsible for — plus the part of the stack you personally owned, named rather than implied.
Software Engineering
Reviewing generated code for correctness rather than style — does it handle the empty input, the second call, the error path — writing reference implementations along with the tests that prove them, specifying rubrics another reviewer can apply consistently, and red-teaming code that sails through the happy path and quietly loses data on the unhappy one.
What tends to count as evidence: Language and domain specificity beats years served. Public repositories, a named stack, and real experience with the kind of code being evaluated — systems, data, mobile, infrastructure — rather than "full-stack" as a label. Expect to demonstrate it rather than assert it.
If you don't have a specialty (yet)
Generalist pools are the normal on-ramp, and they are not a consolation prize — they are where most people learn what this work actually involves: careful reading, consistent judgment, and writing that says precisely what is wrong with an answer. On the board as of October 28, 2026, roles tagged Generalist have a median listed rate of $28/hr across 19 priced listings, with the middle half of them in the $24–$40 band and the extremes running $6–$140.
You may be closer to a specialty than you think
Most people undersell their own credential because they compare themselves to the most senior person in the field instead of to the average applicant. A nurse is a Medicine specialist. A paralegal or contract manager is far closer to the Law pool than to the generalist one. A working translator or interpreter is a Language & Audio specialist with named language pairs, which is precisely the kind of scarce input these pools are built around. A bookkeeper who has closed real books has finance vocabulary a generalist doesn't.
The honest version of the advice: write your profile around the most specific true thing about your working life, not the most impressive one. And qualifications accumulate inside a platform over time — task types often open up as you pass their own checks — so a generalist start does not lock you out of specialist pools later. What nobody can promise you is a number. Anyone telling you what you will earn per hour before you have seen your own matches is guessing.
How to actually target the high-rate pools
Where to go next
- Browse all 317 open roles by specialty →
The listings behind every figure on this page, with the pay each one publishes.
- Run a listed rate through the pay calculator →
Turn an hourly range into a realistic weekly and annual figure, unpaid time included.
- What contractor status means at tax time →
Why a specialist hourly rate isn't a salary, and what that changes about what you owe.