Guide · Cost & scope
Data Annotation Cost: What It Actually Costs in 2026
Annotation pricing is the most opaque corner of AI procurement — most tier-one vendors quote only through sales calls. This guide collects the rates that are actually published, each with its source, and shows how the drivers move a quote.
Published per-unit rates
| Task | Published range | Source |
|---|---|---|
| Image classification | $0.03 – $0.10 per image | BasicAI cost guide |
| Bounding boxes | $0.02 – $1.00 per object | Label Your Data / BasicAI |
| Semantic segmentation | $0.50 – $6.00 per image | HabileData cost guide |
| Medical image segmentation | $2.00 – $8.00 per image | BasicAI cost guide |
| Text annotation | $0.02 – $0.30 per item | HabileData cost guide |
| Audio transcription | $0.50 – $3.00 per minute | HabileData cost guide |
| LiDAR frames | $5 – $25 per frame | HabileData cost guide |
Hourly rates: geography and expertise
| Workforce | Published hourly range | Source |
|---|---|---|
| Africa | $2 – $8 | SecondTalent regional comparison |
| India | $5 – $15 | SecondTalent |
| Philippines | $5 – $12 | SecondTalent |
| Latin America | $8 – $20 | SecondTalent |
| Eastern Europe | $10 – $25 | SecondTalent |
| Western Europe | $20 – $45 | SecondTalent |
| United States | $25 – $60+ | SecondTalent |
| General crowd (transparent platforms) | from $8 (enforced minimum) | Prolific published pricing |
| Credentialed domain experts | $85 – $200 | CleverX published guidance |
| Medical / legal specialists | $250 – $450 | CleverX published guidance |
The pricing models, decoded
- Per unit — predictable for stable tasks; the trap is acceptance rate (see cost per accepted label, below).
- Per hour — right for variable work (RLHF, segmentation, review); demands throughput reporting or you are buying time, not data.
- Platform usage — Labelbox's meter: $0.10 per Labelbox Unit on Starter, with complex rows (multimodal, geospatial) consuming multiple units. Model your mix first.
- Participant + fee — Prolific's model: your chosen hourly pay plus 42.8% (corporate) on top. Fully published; the benchmark for transparency.
- Fixed project / FTE — managed services; commonly gated behind five-figure minimum contracts at enterprise vendors.
Cost per accepted label: the only honest metric
Quotes are not comparable until they are divided by what survives QA. Crowdsourced work looks cheap per task until redundancy (3–5 workers per item), spam filtering and rework pull the effective rate toward managed-service pricing — without the accountability. In-house labeling shocks in the other direction: a $45,000 annotator salary becomes $60,000+ fully loaded, and in-house labels commonly land at 2–4× a managed vendor's rate. Run a 200–500 item pilot with every finalist, measure acceptance, and compare cost per accepted label. That single discipline prevents most bad annotation contracts.
Frequently asked questions
How much does data annotation cost in 2026?
Published ranges: image classification $0.03–$0.10 per image; bounding boxes $0.02–$1.00 per object; semantic segmentation $0.50–$6.00 per image; audio transcription $0.50–$3.00 per minute; LiDAR $5–$25 per frame. Labor runs $2–$60+ per hour depending on region, and credentialed experts $85–$200 per hour. Most enterprise vendors publish none of this — these ranges come from the sources that do.
Why do annotation quotes vary by 100×?
Because "annotation" covers a $0.03 classification click and a $200/hour physician judgment. Geography moves labor 10×, expertise moves it 10× again, and QA design (redundancy, gold sets, expert review) multiplies the unit price. Any single "average cost of annotation" figure is meaningless without the task, the workforce and the acceptance criteria.
What is cost per accepted label?
The only unit cost that matters: total spend divided by labels that pass your QA. A $0.04 label at 60% acceptance costs $0.067; a $0.08 label at 95% acceptance costs $0.084 — and the second vendor also saved you the rework cycle. Every quote comparison should be normalized to accepted output before price is discussed.
Do any annotation vendors publish real pricing?
Yes, a minority: Labelbox publishes usage pricing ($0.10 per Labelbox Unit on Starter, free tier with 500 units/month). Prolific publishes its full model — participant pay you set (minimum $8/hour) plus a 42.8% corporate platform fee. CleverX publishes expert rate guidance ($85–$200/hour typical). Everyone else in our directory is quote-only, and we say so on their profiles.
Sources
- Data annotation pricing: what buyers actually pay — ypai.ai
- Data annotation cost: pricing models & rates — HabileData
- Prolific pricing — prolific.com
- Labelbox pricing from its documentation — UsagePricing
- Best expert data providers for AI labs (expert rate guidance) — CleverX
Facts on this page were drawn from the public sources above, checked November 2026. If you spot an error, tell us at the corrections address in the footer — we fix sourced errors and note the change.
Last updated: November 8, 2026 · Ranges are published figures from the cited sources, not quotes; confirm current pricing with each vendor.