Data annotation services and what data labeling companies charge for image, video and audio

How much do data annotation services cost?

Published per unit rates run from $0.015 for a keypoint to $5.00 and up for a medical label. The two vendors we found who print a rate card list bounding boxes at $0.02 and $0.03 per object. Aggregated category bands put image classification at $0.02 to $0.15, object detection at $0.05 to $0.90, semantic segmentation at $0.50 to $2.00 and up, audio at $0.50 to $3.00 a minute and video at $1.00 to $10.00 a minute.

Do data annotation companies publish pricing?

Almost none. Of twelve providers we opened directly in September 2026, only two published a numeric rate. Four had no pricing page that resolved at all. One category pricing guide states the position openly: most annotation companies do not publish rates, and sales pages say contact us for a quote. That is why comparing two quotes usually means comparing two different definitions of the same word.

Is it cheaper to outsource data annotation or hire an in-house team?

It depends entirely on utilization, and the category cannot agree. One published worked example puts 20,000 images at $15,000 to $18,000 in house against roughly $2,600 outsourced. Another guide tells buyers to expect outsourcing to cost more than 1.5 times an in house team. Both are right for their own assumption. The crossover sits near $0.09 to $0.25 per unit.

How much does data labeling cost per image?

For simple images, between about $0.02 and $0.15 each, and one published worked example lands at $0.13 per image all in for a 20,000 image project including setup. Complexity is what moves it. Segmentation runs $0.50 to $2.00 and up per image and medical or scientific imaging reaches $2.00 to $8.00, because those tasks take minutes rather than seconds.

What is the difference between data annotation and data labeling?

Commercially, nothing. Vendors use the two words interchangeably and price them identically, which is why the same firm often ranks for both. If anyone draws a line it is that labeling suggests assigning a class to a whole item while annotation suggests marking regions, boxes, keypoints or spans inside it. Buy on the specific task and quality bar, never on the label.

How do you choose a data annotation company?

Price the task, not the category. Send the same 200 item sample to every vendor, require an all inclusive quote covering annotation, QA, project management, tooling, rework and reporting, ask for the delivery country and the annotator pay basis, and set a measured agreement threshold against your own gold set. The vendor who scores best on your sample is the answer.

How much should I pay per hour for data annotation?

The third party directory median across 345 annotation companies is $37 an hour, while the cheapest published vendor rate is $6. Both are real and they describe different labor markets. Work out which one your per unit quote implies by multiplying the rate by realistic throughput: at 200 to 400 boxes an hour, a $0.02 box implies $4 to $8 an hour of labor.

Is outsourced data annotation secure enough for regulated data?

Only if you contract for it. Annotation is one of the few AI workflows where humans see raw records one at a time, so treat it as data processing rather than as a service purchase. Require the delivery country in writing, named certifications rather than the word compliant, no personal devices, no offshore subcontracting without consent, and deletion of the working copy on completion.

How long does data annotation take?

Published benchmarks put an experienced annotator at 200 to 400 bounding boxes an hour on clean scenes, dropping to 50 to 150 an hour on dense scenes or where every item is double checked. Individual output runs roughly 150 to 300 images a day against 1,000 to 1,500 for a five person team. Pre annotation with a baseline model roughly doubles throughput.

When should I hire a dedicated annotation team instead of buying per label?

When your per unit price is above roughly $0.09 to $0.25, which is what an hourly annotator costs divided by realistic throughput. Complex, sensitive or continuous work crosses that line: segmentation, medical imaging, RLHF and preference data, anything needing domain knowledge or a stable team that learns your guidelines. Simple bulk images almost never do.

What should be in a data annotation contract?

A measurable quality threshold on an agreed gold set, rework priced at zero below that threshold, the delivery country and subcontracting rule, ownership of the labels and the annotation guidelines you paid to develop, an export path in an open format rather than the vendor platform only, and a stated policy on whether your data may be reused to train anything.

Do I pay for rework and quality assurance separately?

Often yes, and it is the single most common reason a cheap headline rate ends up expensive. A $0.02 box that needs a second pass at $0.02 is a $0.04 box. Ask every vendor to state whether the quoted rate includes QA, how many review passes it buys, what agreement level it guarantees, and who pays when a batch fails that level.