Key findings

1
On the 15 platforms tracked throughout the window, open roles changed by +4.6%, on 937 new listings against 867 removals; 4 grew and 6 shrank by 10% or more while 4 held steady.
Figure 1
2
Among platforms with at least 10 open roles at the start, Mercor grew most like for like, in proportion and in roles (from 206 to 324 open roles, +57.3%).
Figure 1
3
Handshake AI shrank most, in proportion and in roles (from 52 to 0 open roles, -100.0%), with 52 removals against 0 new listings.
Figure 1
4
Over the longer run, the 15 platforms tracked since the week of 3 Aug 2026 averaged 1,476 open roles that week and 1,631 in the week of 31 Aug 2026 (+10.5%).
Figure 3
5
Since 3 Aug 2026, Mindrift moved into AI Evaluation & Rating (RLHF) (13 new listings from 4 Aug 2026, none open at the end) and Handshake AI left Law, Medicine & Finance (41 open roles at the start, none at the end), among 2 entries and 4 exits of at least 5 roles.
Figure 5

Figure 1

Which platforms are growing, holding or shrinking

Over the window, which AI training platforms added open roles, which held and which cut back, and what do they advertise now?

Platform	Open at start	Open at end	Change as read	Change	Status	New listings	Removed	Net flow	Re-key batch	Stating a rate	Median rate
All platforms tracked throughout	1,520	1,590	+4.6%	+4.6%	Steady	937	867	+70	None	89.1%	$65
Invisible	523	523	0.0%	0.0%	Steady	0	0	0	None	99.4%	$65
Mercor	206	324	+57.3%	+57.3%	Growing	351	233	+118	None	97.2%	$85
micro1	277	292	+5.4%	+5.4%	Steady	267	252	+15	None	99.0%	$68
AfterQuery	107	161	+50.5%	+50.5%	Growing	92	38	+54	None	100.0%	$80
Sigma AI	79	79	0.0%	0.0%	Steady	0	0	0	None	0.0%	n/a
Braintrust	73	69	-5.5%	-5.5%	Steady	54	58	-4	None	79.7%	$35
TransPerfect DataForce	42	31	-26.2%	-26.2%	Shrinking	32	43	-11	None	0.0%	n/a
xAI (SpaceXAI)	19	27	+42.1%	+42.1%	Growing	22	14	+8	None	92.6%	$40
Appen	40	24	-40.0%	-40.0%	Shrinking	10	26	-16	None	16.7%	n/a
TeracSmall sample	7	18	+157.1%	+157.1%	Growing	32	21	+11	None	38.9%	n/a
Alignerr	20	17	-15.0%	-15.0%	Shrinking	2	5	-3	None	100.0%	$45
Mindrift	15	13	-13.3%	-13.3%	Shrinking	67	69	-2	None	100.0%	$50
Outlier (Scale AI)	57	9	-84.2%	-84.2%	Shrinking	8	56	-48	None	100.0%	$25
RemotasksSmall sample	3	3	0.0%	0.0%	Too few roles	0	0	0	None	66.7%	n/a
Handshake AI	52	0	-100.0%	-100.0%	Shrinking	0	52	-52	None	n/a	n/a

Sample. 1,590 roles open at the end of 31 Aug 2026 on the 15 platforms with open roles at the start or the end of 1 Aug 2026 to 31 Aug 2026, read from their public job boards; 15 were tracked from before the window opened.

Open at start: open at 00:00 UTC on 1 Aug 2026; open at end: open at the end of 31 Aug 2026. Change compares the two on the same platform, with the start restated by the net of any re-key batch (see Re-key batch), so a re-identification of a board does not read as hiring; Change as read is the raw comparison.

Status: Growing at a change of +10% or more, Shrinking at -10% or less, Steady between. A platform with fewer than 10 open roles at both ends shows Too few roles; a platform first read during the window shows when it was first read and is left out of the first row, so its arrival is never counted as growth.

Re-key batch: a day holding at least 40% of a platform's removals and 40% of its new listings in a period (this window, or a calendar month since 3 Aug 2026), with at least 10 of each. Its listings and removals are set aside from New listings, Removed and Net flow and stated here.

New listings leave out a platform's first week in this dataset (its back catalogue). Net flow is new listings minus removals; it can differ slightly from the change in open roles where a platform's back-catalogue week overlaps the start of the window.

Stating a rate and Median rate are read on the roles open at the end (the first row: every platform with open roles). A median needs at least 8 roles stating an hourly rate.

Source. AITraining.jobs Research, AI Training Vendor Health, August 2026 edition.

Figure 2

The market by status

How many platforms are growing, holding or shrinking, and how many open roles does each group hold?

Status	Platforms	Open at start	Open at end	Change	Which
Growing	4	339	530	+56.3%	Mercor, AfterQuery, xAI (SpaceXAI) and Terac
Steady	4	952	963	+1.2%	Invisible, micro1, Sigma AI and Braintrust
Shrinking	6	226	94	-58.4%	TransPerfect DataForce, Appen, Alignerr, Mindrift, Outlier (Scale AI) and Handshake AI
Too few roles	1	3	3	0.0%	Remotasks
First read in window	0	n/a	0	n/a	None

Sample. The 15 platforms of the panel, 1 Aug 2026 to 31 Aug 2026.

Statuses and changes as in the panel: Open at start is restated by the net of any re-key batch, so it can differ from the panel's start as read. Thresholds: +/-10% like for like; at least 10 open roles at one end.

Source. AITraining.jobs Research, AI Training Vendor Health, August 2026 edition.

Figure 3

Open roles by platform, week by week

How has each platform's number of open roles moved, week by week?

Platform	3 Aug	10 Aug	17 Aug	24 Aug	31 Aug	First to last week
All platforms, as read	1,476	1,509	1,591	1,609	1,631	n/a
Platforms tracked since the first week	1,476	1,509	1,591	1,609	1,631	+10.5%
Invisible	523	523	523	523	523	0.0%
Mercor	202	284	336	338	352	+74.3%
micro1	279	275	299	306	292	+4.7%
AfterQuery	110	116	117	124	161	+46.4%
Sigma AI	79	79	79	79	79	0.0%
Braintrust	71	66	66	71	69	-2.8%
TransPerfect DataForce	25	24	31	31	31	+24.0%
xAI (SpaceXAI)	22	33	34	33	31	+40.9%
Appen	34	29	26	25	23	-32.4%
Alignerr	20	20	20	21	21	+5.0%
Mindrift	18	16	13	19	19	+5.6%
Terac	15	18	22	25	18	+20.0%
Outlier (Scale AI)	45	22	20	11	9	-80.0%
RemotasksSmall sample	3	3	3	3	3	0.0%
Handshake AI	30	1	1	1	0	-100.0%

Sample. Average open roles over the days of each ISO week (Monday to Sunday, UTC; counted at 00:00 each day), 3 Aug 2026 to 31 Aug 2026, on 15 platforms.

Columns are weeks, labelled by their first day. A platform's first week counts only the days after it was first read; weeks before it are n/a, and its change is not computed (it shows when it was first read).

"All platforms, as read" includes platforms as they were added to this dataset, so its change is not computed; "Platforms tracked since the first week" holds the set fixed, so an added platform never reads as growth.

The 31 Aug column covers 1 day only.

Source. AITraining.jobs Research, AI Training Vendor Health, August 2026 edition.

Figure 4

The rate-change log: what each platform offers in its new listings

Is each platform raising or cutting the hourly rate it advertises when it posts new roles?

Platform	Aug 2026 (from 3 Aug)	Median, first to last	Same work	Direction	Compared	Priced new listings
All platforms	$75	n/a	n/a	n/a	n/a	793
Mercor	$85	n/a	n/a	n/a	n/a	308
micro1	$70	n/a	n/a	n/a	n/a	253
AfterQuery	$88	n/a	n/a	n/a	n/a	92
Mindrift	$50	n/a	n/a	n/a	n/a	66
Braintrust	$35	n/a	n/a	n/a	n/a	36
xAI (SpaceXAI)	$40	n/a	n/a	n/a	n/a	20
Outlier (Scale AI)	$10	n/a	n/a	n/a	n/a	8
TeracSmall sample	$100	n/a	n/a	n/a	n/a	7

Sample. 793 new listings stating an hourly rate, 3 Aug 2026 to 31 Aug 2026, by the calendar month they were first listed. A month's median is shown where a platform has at least 5 of them; n/a otherwise.

New listings only (what a platform offers when it posts), not every open role: a rate set months ago and still listed does not move this log. A platform's first week in this dataset (its back catalogue) and the listings of re-key batches are left out.

Median, first to last: the change between the platform's first and last month with a median. It moves with the mix of work listed as well as with rates.

Same work: for each kind of work with at least 5 priced new listings in both of those months, the change in its median, averaged with the later month's listings as weights. Direction reads it: Raised at +10% or more, Cut at -10% or less, Held between; Not comparable when no kind of work qualifies, or when the kinds compared hold less than 50% of the later month's priced listings (Compared), since most of what was posted then has nothing to compare with. For All platforms the same work is the same kind of work on the same platform, so the row is not moved by which platforms posted.

A row is marked as a small sample when its last month with a median rests on fewer than 8 listings.

Source. AITraining.jobs Research, AI Training Vendor Health, August 2026 edition.

Figure 5

Kinds of work each platform entered or left

Which platforms moved into new kinds of AI training work, and which dropped one?

Platform	Kinds at start	Kinds at end	Entered	Left
Terac	5	6	Research Studies & Surveys (2 new from 6 Aug, 2 open at the end); AI Evaluation & Rating (RLHF) (2 new from 5 Aug, none open at the end); Transcription, Audio & Speech (1 new from 7 Aug, none open at the end); Writing & Content (1 new from 26 Aug, 1 open at the end)	Coding & Software Evaluation (2 open at the start)
xAI (SpaceXAI)	6	4	Data Annotation & Labeling (1 new from 13 Aug, 1 open at the end)	Math & STEM (4 open at the start); Transcription, Audio & Speech (1 open at the start); Law, Medicine & Finance (1 open at the start)
Handshake AI	4	0	None	Law, Medicine & Finance (41 open at the start); Math & STEM (6 open at the start); Coding & Software Evaluation (3 open at the start); Transcription, Audio & Speech (2 open at the start)
TransPerfect DataForce	8	7	Writing & Content (1 new from 4 Aug, 1 open at the end)	AI Evaluation & Rating (RLHF) (7 open at the start); Coding & Software Evaluation (3 open at the start)
Mindrift	3	3	AI Evaluation & Rating (RLHF) (13 new from 4 Aug, none open at the end); Math & STEM (1 new from 21 Aug, none open at the end); Writing & Content (1 new from 26 Aug, none open at the end)	None
Outlier (Scale AI)	6	3	None	Coding & Software Evaluation (5 open at the start); Writing & Content (2 open at the start); Law, Medicine & Finance (1 open at the start)
Mercor	9	10	Agentic Tasks & RL Environments (5 new from 26 Aug, 5 open at the end)	None
AfterQuery	8	9	Multilingual & Localization (4 new from 28 Aug, 4 open at the end)	None
micro1	10	9	None	Red-Teaming & AI Safety (2 open at the start)
Braintrust	9	8	None	Research Studies & Surveys (4 open at the start)
Alignerr	3	4	Transcription, Audio & Speech (1 new from 20 Aug, 1 open at the end)	None
Invisible	7	7	None	None
Appen	5	5	None	None
Remotasks	3	3	None	None
Sigma AI	2	2	None	None

Sample. Every platform with open roles at the start or the end of 3 Aug 2026 to 31 Aug 2026, by the kinds of work of its open roles and its new listings.

Entered: no open role in the kind of work at the start of the period (for a platform first read during it, none in its first week's back catalogue) and at least one new listing in it during the period. Left: open roles in the kind at the start and none at the end.

Counts are shown so a move of one or two roles is not mistaken for a strategy: findings rank only entries of at least 5 new listings and exits of at least 5 open roles.

A role's kind of work is its classification today, so a change in this site's classification can move roles between kinds. Listings from re-key batches are not counted as entries.

Source. AITraining.jobs Research, AI Training Vendor Health, August 2026 edition.

Figure 6

Re-key batches set aside

On which days did a platform's listings leave and arrive in one batch, and how large was it?

No data in this window.

Sample. Every platform-day from 3 Aug 2026 to 31 Aug 2026 that meets the re-key rule, in this report's window or in a calendar month of that span.

A re-key day holds at least 40% of a platform's removals and 40% of its new listings in the period read, with at least 10 of each. Ordinary hiring arrives and leaves spread over many days; a single read moving most of both is a change in how the board is identified (new identifiers, or a re-reading of the board by this dataset), so it is reported here and not read as churn.

Same title on both sides: arrivals whose title repeats a departure of the same day on the same platform, matched one to one. Departures that never come back and arrivals with new titles are part of the same batch.

Share columns are against the period read (shown).

Source. AITraining.jobs Research, AI Training Vendor Health, August 2026 edition.

For labs, data vendors and researchers

Buy the data behind this report

The figures above are a summary. Behind them are records we collect ourselves, every day:

10,289 roles tracked, each with its history
Every role on 29 job boards, read daily since 13 Jul 2026: its pay and place, and the first and last day it was open.
6,200 apply clicks in 28 days
Which roles people actually apply to, by field, pay and kind of work. It happens on this site, so nobody else can count it.
39 lab and vendor hiring boards
The job boards of 25 AI labs and 14 human-data vendors, read every day: who is hiring for human data, and who is hiring to buy it.

What you can buy

The records
The rows behind any figure, with their history, not only the totals.
A regular feed
The same figures every week or month, as files.
A custom cut or benchmark
Your countries, languages or fields, or your own platform against the market.

Method and definitions

The public job boards of the AI training platforms this site tracks, read every day since 3 Aug 2026. Listings only: applicant activity on this site is never published per platform. Every figure is computed by one database query over the window shown, the same query for the live report and for every edition. Each figure states its sample beneath it.

Coverage: The platforms whose public job boards AITraining.jobs reads in full every day. Each figure states how many were read.
Open role: A role listed by a given day and still in its platform's feed that day.
New listing: A role that first appeared during the period on a platform already tracked before that week. A platform's first week in this dataset is its back catalogue and is not counted as new.
Removed listing: A role that left its platform's feed during the period.
Net flow: New listings minus removed listings, both outside re-key batches.
Like-for-like change: A platform's open roles at the end of the window against its open roles at the start, for platforms tracked before the window opened. A platform first read during the window is marked and never counted as growth. The start is restated by the net of any re-key batch in the window.
Status: Growing: a like-for-like change of +10% or more. Shrinking: -10% or less. Steady: in between. Too few roles: fewer than 10 open roles at both the start and the end, so no direction is read. First read in window: the platform joined this dataset during the window.
Re-key batch: A single UTC day holding at least 40% of a platform's removals and 40% of its new listings in a period (the report's window, or a calendar month of the last 26 weeks, from 3 Aug 2026 at the earliest), with at least 10 of each. Ordinary hiring arrives and leaves over many days; one read moving most of both is a change in how a board is identified (new identifiers, or a re-reading of the board by this dataset), so the batch is set aside from new listings, removals, net flow and the status, and listed with how many arrivals repeat a departing title.
Advertised pay: The hourly figure a posting states (the middle of a stated range). A platform's median of open roles needs at least 8 roles stating a rate.
Rate-change log: The median advertised hourly rate of a platform's new listings in each calendar month, shown where it has at least 5 new listings stating a rate. Back-catalogue roles and re-key batches are left out.
Same-work change: For each kind of work with at least 5 priced new listings in both the first and the last month with a median, the change in its median, averaged with the later month's listings as weights. Raised at +10% or more, Cut at -10% or less, Held between; Not comparable when no kind qualifies or the kinds compared hold under half of the later month's priced listings. For the whole market the same work is the same kind of work on the same platform.
Entered or left a kind of work: Entered: no open role in the kind at the start of the period (for a platform first read during it, none in its back catalogue) and at least one new listing in it during the period. Left: open roles in the kind at the start and none at the end.
Kind of work: Each role is classified from its title and description into one of the kinds of AI training work this site tracks, from AI evaluation and coding to multilingual and audio work. Every kind links to its own page.
Window: Live: the last 28 complete days, in UTC, with the weekly, monthly and kind-of-work figures read over the 26 weeks to the same end (from 3 Aug 2026 at the earliest). Edition: one calendar month, with those figures read over the 26 weeks to the month's end (from 3 Aug 2026 at the earliest).

What this data cannot tell you

Platforms that recruit without a public job board are not counted. Several large ones invite workers to projects directly.
A listing is not a hire. Some platforms keep evergreen roles open for months, and some post one listing per language or country, so a platform posting per language can grow its listings faster than its hiring.
Open roles at the start and the end are counts on two days: a platform that opens and closes roles in large batches can change status from one window to the next.
A re-key batch is recognised by its shape, not by its cause. A platform that genuinely closed most of its roles and opened as many others on the same day would be set aside too; every batch is listed with its counts so a reader can judge.
Rates are what postings advertise, not what workers are paid after tiers, bonuses or rejected tasks. A month's median of 5 to 7 listings moves with a few postings, and the same-work comparison covers only the kinds of work a platform listed often enough in both months.
A role's kind of work is its classification today, so a change in this site's classification can move roles between kinds.

How the job feeds are read and roles are classified: site methodology. How fresh, complete and consistent the data is, checked every day: data quality and coverage.

Editions

The live report covers the last 28 complete days and changes every day. Each complete month is also published as an edition, frozen two days after the month ends, so a figure you cite from an edition never moves.

Live dataLast 28 complete days, recomputed daily
September 2026 editionFrozen 11 Oct 2026
August 2026 edition (this page)Frozen 11 Oct 2026

Cite and reuse

Free to quote, chart and republish under CC BY 4.0 with a link to this page. This edition is frozen, so its figures will read the same later.

Citation

AITraining.jobs (2026). AI Training Vendor Health, August 2026 edition. https://www.aitraining.run/research/vendor-health-panel/2026-08

BibTeX

@misc{aitrainingjobs_vendor_health_panel_2026_08,
  author = {{AITraining.jobs}},
  title = {AI Training Vendor Health, August 2026 edition},
  year = {2026},
  howpublished = {\url{https://www.aitraining.run/research/vendor-health-panel/2026-08}},
}

Data: every figure as CSV (one line per value) and JSON, sent to your work email.Pay and open-roles data

More research

All 78 research reports