Hiring radar
The AI-training job market, role by role, every day.
Every role that 55 AI-training platforms have posted on their own job boards since 13 Jul 2026: what it pays, which experts it wants, where it is open, when it opened and when it closed. 10,514 roles so far, read every hour, delivered to your team as files.
$25,000 a year, every file below, rebuilt every day.
Who it is for
Read the demand behind AI training before it reaches a report
Investors in the listed AI-data vendors
Contractor hiring at Appen, Innodata, TELUS and RWS is a read on the demand their customers place with them, weeks before it reaches a quarterly report.
Private-market investors and secondary buyers
Mercor, micro1, Invisible, Alignerr, Handshake AI, Mindrift, Surge and Scale's programs: how much expert work each is buying, in which fields, at what pay, and how fast it fills.
Strategy and corporate development teams
Who is hiring which experts for AI training, where, and at what price, measured the same way across every platform.
Data points
Exactly what we hold, and how much of it
Counted from the files themselves, 26 Oct 2026. A subscriber's dictionary file carries the same counts for every column of every file.
| Data point | What it holds | Roles with it | Share |
|---|---|---|---|
| Which role | |||
| Role idrole_id | Our id for the role. The same in every file and every day's delivery. | 10,514 | 100% |
| Platformplatform | The platform that posted the role (our id for it; names in platforms.csv). | 10,514 | 100% |
| The platform's posting idposting_id | The platform's own id for the posting, as its job board gives it. | 10,514 | 100% |
| Titletitle | The title as the platform posted it. | 10,514 | 100% |
| Client named by the platformhiring_org | The organisation the platform names as the client the role is for, when it is not the platform itself. | 105 | 1% |
| What work | |||
| Kind of workkind_of_work | Our classification of the work into one of 12 kinds (for example coding, domain-expert, multilingual), the same for every platform. | 10,514 | 100% |
| The platform's own fieldplatform_field | The platform's own department, field or domain label for the role, where its board gives one. | 7,672 | 73% |
| Kind of contractcontract | The kind of engagement as the platform wrote it (contract, freelance, part-time, hourly). | 7,981 | 75.9% |
| Pay | |||
| Advertised pay, as writtenpay_text | The pay line exactly as the platform advertised it. | 9,217 | 87.7% |
| Hourly pay, low endpay_hourly_low | The lower end of the advertised hourly pay in USD, read from pay_text. Hourly rates only: salaries and per-task pay stay in pay_text. | 8,899 | 84.6% |
| Hourly pay, high endpay_hourly_high | The upper end of the advertised hourly pay in USD (equal to the lower end for a single rate). | 8,899 | 84.6% |
| Hourly pay, one figurepay_hourly | One hourly figure for the role: the middle of its range, or its single rate. | 8,899 | 84.6% |
| Where | |||
| Locationlocation | The location as the platform wrote it. | 9,262 | 88.1% |
| Remote flagremote | true when the platform marks the role remote. | 10,514 | 100% |
| Who can take it, by countryopen_to | Who can take the role by country: global (anywhere), include (only the countries listed), exclude (anywhere but those), unknown (the posting does not say). | 9,930 | 94.4% |
| Countries namedcountries | ISO country codes the role is limited to (open_to include) or closed to (open_to exclude), separated by spaces. | 6,329 | 60.2% |
| When | |||
| The platform's posting dateposted_at | When the platform says it posted the role (UTC), where its board gives a date. | 8,430 | 80.2% |
| First recorded by usfirst_recorded_at | When our read first recorded the role (UTC). A platform's roles in its first week in our data were already listed when we started reading it. | 10,514 | 100% |
| Current listing sincelisted_since | When the role's current listing began: equal to first_recorded_at unless the role was gone for over a week and came back. | 10,514 | 100% |
| Last seenlast_seen_at | The last read that listed the role (UTC). For an open role, the latest read. | 10,514 | 100% |
| Open or closedstatus | open when the latest read listed the role, closed when it did not. | 10,514 | 100% |
| Days listeddays_listed | Days from posted_at (or first_recorded_at where the board gives no date) to last_seen_at. | 10,514 | 100% |
| Days listed counted fromdays_listed_from | Which date days_listed counts from: posted or recorded. | 10,514 | 100% |
- Role records
- 10,514 roles since 13 Jul 2026: 3,896 open at the latest read, 6,618 closed. 8,311 distinct titles, 232 countries named, 8,899 with an hourly rate, and 241,476 role-days observed.
- Platform by day
- 2,085 platform-days since 13 Jul 2026: open, new and closed roles and the median advertised hourly pay, for each of the 55 platforms on every day since we first read it.
- Market by day, as recorded
- 52 days recorded from 29 Aug 2026 to 26 Oct 2026, each with the market as a whole and each of the 12 kinds of work. 7 days in that span were not recorded: 7 Sep 2026, 8 Sep 2026, 9 Sep 2026, 11 Sep 2026, 12 Sep 2026, 14 Sep 2026, 15 Sep 2026.
- Change log
- 58 changes since 26 Oct 2026: 58 closed, 0 reopened, 0 pay moves, 0 pay lines reworded and 0 titles changed, each with the time of the read that saw it.
- Reconstructed history, before our own reads
- 15,185 AI-training postings seen in 572 archived captures of the platforms' own boards, or on their own archived posting pages, the earliest from 8 Jan 2024. Each capture is a public page at the Internet Archive anyone can open, and it is kept apart from what we observed ourselves.
- Mercor4,989 postings (3,032 from posting pages), from 30 Mar 2025
- Alignerr3,194 postings, from 15 Aug 2025
- Welo Data1,695 postings, from 12 Nov 2024
- micro11,583 postings (1,583 from posting pages), from 4 Nov 2024
- Invisible776 postings (1 from posting pages), from 23 Jan 2025
- Appen733 postings, from 23 Mar 2024
- Outlier (Scale AI)427 postings (357 from posting pages), from 6 Nov 2024
- RWS TrainAI Community363 postings, from 7 Nov 2024
- OneForma283 postings (136 from posting pages), from 8 Jan 2024
- Sigma AI241 postings, from 13 Jun 2024
- DataAnnotation224 postings (96 from posting pages), from 2 Apr 2026
- Rise Data Labs97 postings, from 21 Aug 2026
Counts are each board's own posting ids, not merged into roles the way our own records are: a board that posts one role many times (Alignerr does) holds many ids for it. Where a capture shows only a board's first page, its stated total is in the file beside the sample.
Coverage
55 platforms, each from the day we first read it
How long we have read each platform and how complete its fields are. The share columns are the percent of its roles that carry the field.
| Platform | First read | Roles | Open now | Hourly pay | Posting date | Countries |
|---|---|---|---|---|---|---|
| OpenTrain AI | 5 Sep 2026 | 1,886 | 41 | 82.3% | 100% | 99.9% |
| Mercor | 13 Jul 2026 | 1,158 | 237 | 91.6% | 100% | 99.8% |
| micro1 | 13 Jul 2026 | 1,046 | 270 | 96.9% | 0% | 100% |
| Mindrift | 13 Jul 2026 | 1,025 | 56 | 99.9% | 100% | 71.8% |
| Alignerr | 13 Jul 2026 | 870 | 377 | 100% | 0% | 81.4% |
| Invisible | 13 Jul 2026 | 807 | 534 | 98.5% | 100% | 97.1% |
| SuperAnnotate (SME) | 10 Oct 2026 | 391 | 381 | 100% | 100% | 100% |
| Welo Data | 8 Sep 2026 | 364 | 182 | 70.3% | 100% | 100% |
| Weekday | 18 Oct 2026 | 333 | 168 | 89.2% | 100% | 100% |
| Prolific | 13 Jul 2026 | 299 | 124 | 61.5% | 100% | 81.9% |
| Braintrust | 13 Jul 2026 | 248 | 77 | 82.3% | 100% | 100% |
| Handshake AI | 13 Jul 2026 | 227 | 165 | 99.6% | 100% | 100% |
| TransPerfect DataForce | 13 Jul 2026 | 211 | 57 | 3.3% | 100% | 100% |
| Anyone AI | 11 Oct 2026 | 134 | 132 | 99.3% | 100% | 100% |
| RWS TrainAI CommunityRWS (AIM: RWS) | 22 Oct 2026 | 130 | 127 | 91.5% | 100% | 89.2% |
| Sigma AI | 13 Jul 2026 | 130 | 91 | 0% | 100% | 95.4% |
| AppenAppen (ASX: APX) | 13 Jul 2026 | 118 | 54 | 17.8% | 100% | 83.1% |
| DataAnnotation | 22 Oct 2026 | 112 | 112 | 100% | 0% | 100% |
| Rise Data Labs | 18 Oct 2026 | 107 | 106 | 75.7% | 100% | 99.1% |
| xAI (SpaceXAI) | 26 Jul 2026 | 87 | 14 | 92% | 100% | 100% |
The other 35 platforms
| Terac | 13 Jul 2026 | 82 | 32 | 36.6% | 100% | 100% |
|---|---|---|---|---|---|---|
| Outlier (Scale AI) | 13 Jul 2026 | 78 | 5 | 100% | 100% | 100% |
| Volga Partners | 18 Oct 2026 | 75 | 49 | 80% | 100% | 100% |
| TELUS Digital AITELUS (TSX: T, NYSE: TU) | 22 Oct 2026 | 71 | 66 | 73.2% | 100% | 100% |
| Perle Labs | 10 Oct 2026 | 67 | 67 | 0% | 100% | 100% |
| Claru | 14 Oct 2026 | 60 | 60 | 100% | 100% | 100% |
| Lifted (Upwork) | 18 Oct 2026 | 56 | 38 | 16.1% | 100% | 100% |
| Babel Audio | 11 Oct 2026 | 43 | 43 | 53.5% | 100% | 100% |
| RemoExperts | 10 Oct 2026 | 39 | 38 | 94.9% | 100% | 100% |
| InnodataInnodata (Nasdaq: INOD) | 18 Oct 2026 | 38 | 31 | 76.3% | 100% | 100% |
| OneForma | 14 Oct 2026 | 37 | 19 | 2.7% | 100% | 100% |
| Sovrano AI | 11 Oct 2026 | 27 | 18 | 85.2% | 100% | 100% |
| iMerit | 22 Oct 2026 | 22 | 9 | 63.6% | 0% | 100% |
| Vetto | 14 Oct 2026 | 20 | 20 | 15% | 100% | 100% |
| Aligned Labs | 22 Oct 2026 | 15 | 15 | 100% | 100% | 100% |
| Surge AI | 22 Oct 2026 | 13 | 13 | 92.3% | 0% | 100% |
| Toloka | 13 Jul 2026 | 13 | 0 | 0% | 100% | 15.4% |
| DATAmundi | 22 Oct 2026 | 11 | 9 | 0% | 100% | 100% |
| Scale AI Human Frontier Collective | 18 Oct 2026 | 11 | 9 | 27.3% | 100% | 100% |
| Shaip (Collaborate by Shaip) | 22 Oct 2026 | 10 | 8 | 10% | 0% | 100% |
| Pasiflora AI | 14 Oct 2026 | 7 | 7 | 42.9% | 0% | 100% |
| Elsevier HealthRELX (LSE: REL, NYSE: RELX) | 25 Oct 2026 | 6 | 6 | 100% | 100% | 100% |
| Pila8 | 11 Oct 2026 | 4 | 4 | 75% | 100% | 100% |
| RealTrainAI | 22 Oct 2026 | 4 | 4 | 0% | 0% | 100% |
| Remotasks | 13 Jul 2026 | 4 | 3 | 75% | 100% | 75% |
| Cohere | 25 Oct 2026 | 3 | 3 | 0% | 100% | 100% |
| Mistral AI | 25 Oct 2026 | 3 | 3 | 0% | 100% | 100% |
| Sama | 22 Oct 2026 | 3 | 3 | 33.3% | 100% | 100% |
| Anthropic | 21 Oct 2026 | 2 | 2 | 0% | 100% | 100% |
| Fleet AI | 22 Oct 2026 | 2 | 2 | 100% | 100% | 100% |
| Headway | 25 Oct 2026 | 1 | 1 | 100% | 100% | 100% |
| Limbic | 25 Oct 2026 | 1 | 1 | 0% | 100% | 100% |
| PointClickCare | 25 Oct 2026 | 1 | 1 | 100% | 100% | 100% |
| Snorkel AI Expert Community | 22 Oct 2026 | 1 | 1 | 100% | 100% | 100% |
| Vals AI | 25 Oct 2026 | 1 | 1 | 100% | 100% | 100% |
What you receive
Everything in one subscription
$25,000 a year for your firm: every file below, rebuilt every day, at addresses that are yours for the year.
Role records
roles.csvOne row per AI-training role we have recorded since our first read, open or closed: platform, the platform's own posting id, title, kind of work, the platform's own field, advertised pay as written and as an hourly range, where it is open, remote or on site, contract type, when the platform posted it, when we first and last saw it, and whether it is still open. The whole file is rebuilt every day.
10,514 rows today
Change log
changes.csvEvery change our reads see on a role, with the time of the read: it closed, it reopened, its advertised pay moved (old and new, as written and as an hourly range), its title changed.
58 rows today
Platform by day
platform-days.csvFor every platform and every day since we first read it: roles open, roles that appeared, roles that closed, and the median advertised hourly pay of the open roles.
2,085 rows today
Market by day, as recorded
market-days.csvThe market as we recorded it that day: open roles, roles new in the week, platforms hiring, and per kind of work the open roles, the median and typical advertised hourly pay and how many rates they rest on.
666 rows today
Coverage by platform
platforms.csvFor each platform: the day we first read it, how many roles we hold, how many are open, and how complete each field is.
55 rows today
Reconstructed history
history-postings.csv, history-captures.csvThe platforms' own job boards as the Internet Archive captured them, from 2024 where captures exist, rebuilt posting by posting and kept apart from what we observed ourselves, so a reader always knows which is which.
15,185 rows today
Data dictionary and method
dictionary.csvEvery field of every file: what it means, how it is measured, how many rows carry it and since when.
95 rows today
- Files as CSV, or JSON with ?format=json, at addresses that are yours for the year: open them in a spreadsheet or fetch them from code every day.
- The files are rebuilt from the job boards we read every hour; each file states when it was built.
- Licensed to your firm for its own research and investment work, for twelve months.
Method
How it is measured
- We read each platform's own public job board, its applicant tracking system's feed or its own listings, every hour. Each platform's first day in our data is in the coverage table.
- Only roles read from job boards anyone can open without signing in and without passing a check against automated reading are sold, role by role. A platform read any other way stays on our own site and in the market's daily totals as recorded, never in the role files.
- One row is one role. A role a platform posts once per country, or reworded several times at the same pay, is kept once, with all its countries.
- Pay is the platform's own advertised line, kept as written. Hourly rates and ranges are also read into USD figures; salaries and per-task pay stay as written.
- A platform's roles in its first week with us were already listed when we started reading it, so they are never counted as newly opened (the first_week column of the platform-by-day file).
- A role closes on the first read that no longer lists it; last_seen_at is the last read that did.
- Until 26 Oct 2026 a role kept only its latest state: its pay before a change, and the days a reopened role was away, were not kept. From then on the change log has every one of them.
- Rebuilt from our role records, the open-role count matches the count recorded on the day within a median 4.2% across 52 recorded days. The largest gap, 15.2% on 17 Sep 2026, comes from roles counted that day that are no longer in our role records; the market-by-day file shows both counts side by side for every day.
- Kinds of work are our own classification, the same for every platform: AI Evaluation & Rating (RLHF), Coding & Software Evaluation, Math & STEM, Law, Medicine & Finance, Agentic Tasks & RL Environments, Data Annotation & Labeling, Robotics & Physical-AI Data, Multilingual & Localization, Transcription, Audio & Speech, Writing & Content, Red-Teaming & AI Safety, Research Studies & Surveys.
Free to read
What stays free
Our research reports and a monthly aggregate snapshot stay free to read and cite. Hiring radar is what they are built from: each role, each day, each change.
Subscription
$25,000
a year, for your firm
Every role open on any day of September 2026 with all its fields, every platform's days in September, the market's days in September, and the first week of the change log. The same files and fields a subscriber receives, sent to your work email.