Data annotation pricing
With Data annotation pricing, you're charged by the label, by the item or by the hour. Pick a label type to find your project's cost; published offshore and US rates apply, and each is linked to where it appeared.
Work out the cost of your labeling job
50,000 boxes overall, drawn as squares of 1,000 boxes
That makes 50,000 boxes across 10,000 images. What a team outside the US would charge, roughly:
$1,000 to $6,000
Teams abroad list between $0.02 and $0.12 for each box.
Get quotes for this jobThe calculation behind this
The published low and high price for that label type gets multiplied by the label count. Another review adds 20 to 50 percent, while a US team costs two to four times the outside-US range.
Prices for image tags come from Kili Technology for the low end and BasicAI for the high end.
Prices for bounding boxes come from Label Your Data for the low end and Habile Data for the high end.
Prices for polygons come from AnnotationBox for the low end and Habile Data for the high end.
Prices for keypoints come from Label Your Data for the low end and Habile Data for the high end.
Prices for pixel masks come from Habile Data for the low end and GDS Online for the high end.
Prices for video tracking come from Habile Data.
Prices for text entities come from Habile Data.
Prices for sentiment come from GDS Online.
Prices for transcripts come from GDS Online for the low end and Habile Data for the high end.
Prices for lidar cuboids come from Habile Data.
Prices for answer rankings come from Habile Data.
Published rates by label type
What teams outside the US charge for every kind of label, as published. As an estimate, the US column multiplies those prices by two to four, the markup Habile Data says applies to US teams.
| Label | Price outside the US | Unit | Estimated US price (2 to 4 times) | Source (low / high) | Other published prices |
|---|---|---|---|---|---|
| Image tags | $0.01 to $0.10 | per image | $0.02 to $0.40 | low: Kili Technology; high: BasicAI | Habile Data: $0.05 per image; Mindkosh: $0.035 per image |
| Bounding boxes | $0.02 to $0.12 | per box | $0.04 to $0.48 | low: Label Your Data; high: Habile Data | AnnotationBox: $0.04 per box; Annotation Support: $0.036 per label |
| Polygons | $0.06 to $0.80 | per polygon | $0.12 to $3.20 | low: AnnotationBox; high: Habile Data | Mindkosh: $0.07 per polygon (≤ 8 points); US Data ML: from $0.15 per polygon |
| Keypoints | $0.015 to $0.10 | per point | $0.03 to $0.40 | low: Label Your Data; high: Habile Data | Annotation Support: $0.021 per keypoint; Mindkosh: $0.025 per point |
| Pixel masks | $0.20 to $3.00 | per image | $0.40 to $12.00 | low: Habile Data; high: GDS Online | Mindkosh: $3 per image; Kili Technology: $0.10 to $1.00 per mask outside the range |
| Video tracking | $0.05 to $0.25 | per tracked object per frame | $0.10 to $1.00 | Habile Data | GDS Online: $0.10 – $0.50+ per frame outside the range |
| Text entities | $0.04 to $0.30 | per document | $0.08 to $1.20 | Habile Data | Label Studio: $0.02 per entity outside the range; Kili Technology: $0.024 to $0.70 per unit of text outside the range; US Data ML: from $0.25 per text |
| Sentiment | $0.01 to $0.15 | per text | $0.02 to $0.60 | GDS Online | Kili Technology: $0.001 to $0.13 per unit of text outside the range; US Data ML: from $0.06 per text |
| Transcripts | $0.10 to $1.50 | per audio minute | $0.20 to $6.00 | low: GDS Online; high: Habile Data | Kili Technology: $0.10 to $10.00 per minute of audio outside the range; US Data ML: from $0.25 per minute |
| LiDAR cuboids | $0.10 to $0.50 | per cuboid | $0.20 to $2.00 | Habile Data | Annotation Support: $0.121 per cuboid; Mindkosh: $0.2 per cuboid; BasicAI: $3.00 per label outside the range |
| Answer rankings | $0.50 to $8.00 | per comparison | Priced by the expert, not the location | Habile Data | Grading code or medical answers and similar expert tasks: $20 to $100 each, or higher. |
Column two holds the range the calculator uses; "Source (low / high)" links pages that set its ends. A figure tagged "outside the range" is another published price; it usually reflects a different unit, "per unit of text" for example, or premium work. Ask for a quote on a sample of your data and you'll see where your job falls.
Three ways vendors charge for labeling
With per-label pricing, you pay for every point, polygon or box. This fits object detection, since the label count is easy to predict there.
With per-item pricing, you pay for every audio minute, document or image. Tags, masks and transcripts fit this model.
Per hour: annotator time is what you're buying. Label Your Data charges $6 an hour per annotator; onshore generalists are $20 to $60 an hour. Paying by the hour fits jobs you can't easily count, such as lengthy checks of model output.
Things that push the price up
- Team location matters: work delivered from the US runs two to four times what offshore teams charge.
- Accuracy needs: a polygon or mask takes much longer than a box, so each object costs more.
- Quality checks: a second pass or a consensus review adds 20 to 50 percent on top.
- Specialized data: you pay 3 to 5 times more for medical imaging than for ordinary images.
Your own staff versus outsourcing
According to a cost comparison by Habile Data, labeling 1,000 boxes runs $150 to $400 with in-house staff once you count salaries, tools and checking. An offshore vendor charges $40 to $120, slightly above the bare per-box rate because review is included. Keeping it in-house still makes more sense when your data has to stay on your premises.
Common questions about data annotation pricing
What does annotating an image cost?
Expect $0.01 to $0.10 per image for simple tags. For ordinary scenes, pixel masks cost $0.20 to $3.00 per image; dense street scenes or medical scans cost much more.
How much is data annotation per hour?
Generalists offshore charge about $4 to $10 an hour; onshore generalists in the US charge $20 to $60. Expect to pay $50 to $150 or more for domain experts such as clinicians or lawyers.
What makes vendors quote so differently for the same job?
Checking levels, team locations and speed all differ from one vendor to another, and each builds them into the price. Find out how many labels per hour each one plans and what share they review; that lets you compare like with like.
Send one brief, hear from three vendors
Describe your data once and we pass it to three annotation vendors that label that kind of data, so you can compare quotes side by side.
- One short form instead of three sales calls
- Matched on data type, volume and security needs
- You choose who to talk to