Data labeling cost in 2026: pricing per hour and per label

Published prices checked on each company's own website on 24 October 2026.

Several managed data labeling teams publish rates of $5 to $7 per person-hour for common work, and simple labels start at a few cents each: published prices begin at $0.02 per bounding box and $0.02 per text entity. Teams based in the US or Western Europe, and specialist work, cost more and are usually quoted. Deepen AI Services charges $5 per person-hour for human review and data operations, with a trained team, a team lead, QA sampling and a weekly report included. Specialist work, such as LLM evaluation, RLHF and robotics data, and complex labels, such as 3D point clouds, are quoted per project. The number that matters most is the cost of each label you accept, after review and rework.

Published prices at a glance

The five ways data labeling is priced

• Per person-hour. You pay for the time people work. It suits tasks that need judgment, mixed items and rules that change, and review work of any kind. Ask what each hour includes: training, QA, a team lead and reporting.
• Per label or per object. A price for each box, polygon, entity or item. It's easy to budget for simple, uniform labels. On busy images the number of labels drives the bill, so agree up front how labels are counted.
• Participant pay plus a fee. Research panels add a percentage to what you pay participants. It suits surveys and studies, where you design the task and review the answers.
• Software plus your own team. A labeling tool priced per seat. The tool is cheap; hiring, training and reviewing the people is the real cost.
• Custom contracts. Enterprise programs priced per statement of work, often with annual commitments. They suit large, long-running programs.

What changes the price

• The label type. A yes-or-no decision costs less than a box, and a box less than a polygon or a 3D cuboid.
• Judgment and expertise. Rating LLM answers, reading contracts or scoring code needs trained or specialist people, and costs more than tagging photos.
• The quality bar. Review passes, gold sets with known answers and agreement checks add work, and cut the rework you'd pay for later.
• Volume and steadiness. Steady weekly work lets a team stay trained and fast. Stop-start batches cost more per item.
• Turnaround. Same-day or round-the-clock work needs more people or shifts.
• Language. Work outside English, or in several languages, needs people fluent in each.
• Security. In-house teams, an NDA, work inside your own systems and audits such as SOC 2 and ISO 27001 matter for customer data and documents. Ask which a price includes.

The costs a quote leaves out

The rate on a quote is rarely the full cost. Check who pays for each of these:

• Writing and updating the guidelines, and answering questions about them
• Training people before live work, and retraining when the rules change
• Your own team's time reviewing samples and fixing errors
• Redoing rejected labels, and the delay while they're redone
• Tool seats, storage and the work of moving data in and out
• Someone managing the vendor day to day

With a managed team such as Deepen AI Services, training, QA and a team lead are part of the hourly price, and the team works inside your tools, so there's no data to move.

Compare quotes on the cost of each accepted label

The fair way to compare offers is the cost per accepted label: everything you pay, including your own review time and rework, divided by the labels you keep.

Cost per accepted label = (labeling + your review + rework + tools) ÷ labels you accept

How Deepen AI Services pricing works

• The rate: $5 per person-hour for human review and data operations in English, done inside your tools: AI output review, agent action approval, exceptions, moderation, document extraction, text labeling, product data, data entry and transcription.
• What's included: People trained on your guidelines before live work, a named team lead, QA sampling and a weekly report.
• Quoted in your plan: AI evaluation, agent training data and robotics data, work in other languages, and image, video and sensor annotation, which the enterprise team scopes.
• How you start: Describe the task and get a plan and a price by email within 2 working days, with no sales call. Then a two-week pilot on your real work, prepaid at people × hours per week × 2 weeks × $5.
• An example: Two people at the 20-hour weekly minimum cost $400 for the two-week pilot and $200 a week after it, invoiced weekly.

See the full pricing, or get an exact price for your task.

How to bring labeling costs down

• Let AI do the first pass. Pre-label with a model and have people check and correct it. Checking is faster than labeling from scratch.
• Write guidelines with examples. Finished examples and the edge cases you'd reject cut questions and rework. See how to write annotation guidelines.
• Check against known answers. A small gold set catches drift early, before it costs a batch.
• Keep the work steady. A team that works on your task every week stays trained and gets faster.
• Pilot before you scale. Two weeks on real work shows the true rate per item before you commit.
• Label only what the model needs. Skip attributes and classes you'll never use.

Questions to ask before you sign

• What does the rate include: training, QA, a team lead, reporting?
• How are labels or hours counted, and what's the minimum?
• Who reviews the work, how much of it, and how do you see the results?
• What happens to rejected work, and who pays to redo it?
• Is there a pilot, what does it cost, and how do its results carry into production pricing?
• Who are the people: an in-house team or a crowd?
• Where does the data live, and which certifications apply?
• How are you invoiced, and what are the commitments?