Anyone can quote you a number. Few can tell you why the number moves. Here's what drives annotation pricing - and the questions to ask before you commit.
Bounding boxes on images are cheap. 3D cuboids on LiDAR, temporal video tracking, and audio transcription each bring their own effort curve.
Ten classes in clean daylight is not the same project as 200 classes in rain, shadow, and occlusion. The label taxonomy drives the hour.
Double annotation, expert review, and 99%+ agreement targets cost real money. Decide the bar your model actually needs, not the highest one on the menu.
Generic English text is commodity work. Rare dialects, legal or medical writing, and specialized domains command specialist rates.
Bigger committed volumes and patient timelines lower unit cost. Last-minute fire drills raise it. Plan ahead and you pay less per unit.
Some vendors quote per raw item and charge for rejections later. Ask for per accepted unit pricing so you're comparing apples to apples.
Single-pass labeling is cheap because it's cheap. Ask what review depth is included and what agreement rate they guarantee.
Who eats the cost of rejected work? The answer tells you how aligned the vendor is with your quality goals.
Any vendor that won't do a small paid pilot at honest pricing is selling you something you'll regret at volume.
Send your data type, volume, and quality bar - we'll come back with per-unit pricing and a sample batch, no games.
Request a Quote
XYNTRIQ