For teams buying AI training data

Scale AI pricing

Scale AI publishes no price: its pricing page is a form to book an introductory call. The press reports that most of its revenue comes from charging model makers for access to its network of specialised human trainers.

Compare data vendors

Read against Scale AI’s own pages and the press on 2 Nov 2026. Every figure below links its source.

The figures

What Scale AI charges, keeps and pays

Each figure as its source published it, with whose money it is: what a buyer pays, what the vendor keeps, or what the person doing the work receives.

Figure as published	Whose money	Basis	Source
Historically about 2 cents per image and 6 cents per annotation on simple tasks, with multipliers for language, consensus and complexitysimple image labeling (historic unit rates)	Paid by the buyer	Analyst estimate	HeroHunt.ai, 6 Sep 2026
$500 million contract, a five-fold increase on a $100 million deal in September 2025data processing and decision support	Paid by the buyer	Stated by the company	Bloomberg, 6 May 2026
At least $450M per year for five years, as part of the deal for a 49% stakeAI products and data services	Paid by the buyer	Reported by a third party	TheFly via TipRanks (summarising Forbes), 13 Jun 2025
Google had planned to pay Scale AI $200 million in 2025 before moving to cut tieshuman-labeled training data	Paid by the buyer	Reported by a third party	TechCrunch (citing Reuters), 2025
Google spent around $150 million on Scale AI services in 2024human-labeled training data	Paid by the buyer	Reported by a third party	Storyboard18 (citing Reuters), 2024
Monetizes per task, marking up the cost of labor, for 50%+ gross margindata labeling	Kept by the vendor	Analyst estimate	Sacra, 11 Oct 2026
Up to $1,350 per task; tasks typically take 6 to 7 hoursfrontier-difficulty tasks (coding, ML, security, sciences)	Paid to the worker	Stated by the company	Outlier (Scale AI), 11 Oct 2026

From our dataset of published figures on what AI companies pay for human data: every vendor’s figures in one report.

About Scale AI

What Scale AI sells

A data engine for AI labs (expert human feedback, post-training and evaluation data from specialised trainers) and an applications business for governments and enterprises.

Scale AI supplies credentialed experts for AI training work, as its own pages describe.

In June 2025 Meta invested in Scale AI at a valuation of over US$29 billion, and founder Alexandr Wang left to join Meta. Reuters put Scale's 2024 revenue at US$870 million.

Pricing (Scale AI, 2 Nov 2026)
Google, Scale AI's largest customer, plans split after Meta deal (CNBC (Reuters), 14 Jun 2025)

What an expert hour costs

Start from what the expert is paid

Every vendor's price for an expert hour is the expert's pay plus the vendor's share. We read the pay of every open AI-training role on the major platforms every hour, so here is what each kind of US expert is paid for AI work today.

The live rates are unavailable right now.

What other vendors keep

Appen: Gross margin 43.3% in Q3 FY26 (39.2% a year earlier); 39.2% for January to September 2026 (stated by the company, Appen Limited (ASX: APX), ASX announcement, 29 Oct 2026).
Handshake: AI data business gross margins around 30%, versus about 80% for the job board (analyst estimate, Sacra, 11 Oct 2026).
Mercor: Percentage-based recruiting fee, typically 30%, for direct talent placements (analyst estimate, Sacra, 11 Oct 2026).
Mercor, Handshake, micro1: Roughly 60% to 70% of top-line income paid out to domain specialists (reported by a third party, TechCrunch, 22 Sep 2026).
Mercor: Third-party analysts estimate Mercor's take at roughly 30% to 35% of billed hours, with contractors receiving 60% to 70% of gross billings (analyst estimate, HeroHunt.ai, 6 Sep 2026).
Handshake AI: Pays about $60 million a month to more than 30,000 individuals, on a run rate of about $1 billion (stated by the company, Handshake (job board), 5 Sep 2026).
micro1: Gross margins on off-the-shelf data as high as 80% to 90% (reported by a third party, TechCrunch, 20 Aug 2026).
micro1: Net annual run rate $150M to $200M on a $500M gross run rate (about 30% to 40% retained) (reported by a third party, TechCrunch, 20 Aug 2026).

Alternatives

Alternatives to Scale AI

Expert data from US professionals, run end to end

Evaluations, rubrics, fine-tuning data and red-teaming by US doctors, nurses, lawyers, finance professionals, scientists and engineers from our own network. An expert hour is priced at what that kind of expert is paid today (the table above) plus our fee for running the project, and the quote shows both.

How our expert data projects work

By field: Medical · Legal · Finance · Coding · Science and engineering · Red-teaming and safety · Languages

Appen pricingAppen publishes no price list: there is no pricing page on its site, and each project is quoted. Its reports describe revenue from projects with AI model builders and from prebuilt datasets, without unit prices.Shaip pricingShaip publishes no price: its pricing page shows no figure, rate card or minimum, and asks buyers to talk to its team for a quote tailored to the project.Toloka pricingToloka publishes no rate card. On its self-service platform the buyer prepays an account balance and sets the hourly rate it will pay within Toloka's suggested range; fees are shown in the account at checkout, exclude VAT, and purchases are non-refundable. Enterprise projects are quoted.TELUS Digital pricingTELUS Digital publishes no price or pricing model: its AI training data page offers projects by quote.Prolific pricingProlific publishes its prices: buyers pay participants plus a platform fee, usually 42.8% for companies and 33.3% for academic and non-profit customers, with no contract or subscription. Participants must be paid at least US$8 an hour. Managed projects are quoted.Surge AI pricingSurge AI publishes no price: there is no figure anywhere on its site, and projects are scoped with its team.

All 25 data vendors and how each prices

Looking for paid work on Outlier rather than buying data? Outlier (Scale AI): open roles, pay and whether it is legit.

Questions

What teams ask about Scale AI

How much does Scale AI charge?

What have buyers paid Scale AI?

Historically about 2 cents per image and 6 cents per annotation on simple tasks, with multipliers for language, consensus and complexity (analyst estimate, HeroHunt.ai, 6 Sep 2026). $500 million contract, a five-fold increase on a $100 million deal in September 2025 (stated by the company, Bloomberg, 6 May 2026). At least $450M per year for five years, as part of the deal for a 49% stake (reported by a third party, TheFly via TipRanks (summarising Forbes), 13 Jun 2025).

How much of the price does Scale AI keep?

Monetizes per task, marking up the cost of labor, for 50%+ gross margin (analyst estimate, Sacra, 11 Oct 2026).

What does Scale AI sell?

A data engine for AI labs (expert human feedback, post-training and evaluation data from specialised trainers) and an applications business for governments and enterprises.

What should a buyer know about Scale AI as a company?

In June 2025 Meta invested in Scale AI at a valuation of over US$29 billion, and founder Alexandr Wang left to join Meta. Reuters put Scale's 2024 revenue at US$870 million.