Vendor profile · Directory · Platform
Lightly AI
An ETH Zurich spin-off built on one idea: label less, but label the right data — curation and self-supervised pretraining first, annotation second.
| Model | Data curation + pretraining platform (ETH Zurich spin-off) |
|---|---|
| Founded | 2019 |
| Headquarters | Zurich, Switzerland |
| Pricing | Not published — platform and services quoted; open-source tools free |
| Website | www.lightly.ai |
What Lightly AI does
Lightly was founded in Zurich in 2019 by ETH and Harvard alumni Igor Susmelj and Matthias Heller. LightlyOne selects and curates the samples worth labeling from large unlabeled pools; LightlyStudio (open source, Apache 2.0) covers curation and labeling; LightlyTrain pretrains vision models on your own unlabeled data.
The company also offers managed annotation services and RLHF with domain experts, but its center of gravity is computer-vision data efficiency for teams in retail, agriculture, healthcare and manufacturing.
Strengths
- Data selection/curation is the product, not a feature bolt-on
- Open-source Studio lowers the cost of evaluating the approach
- Pretraining (LightlyTrain) attacks the same cost problem from the model side
Considerations
- No published pricing for the commercial platform
- Vision-only focus
- Smaller company — check support and roadmap depth in diligence
Other platforms in the directory
- Labelbox — The platform pick: run annotation in-house on published usage pricing, and bolt on expert services (Alignerr) only where you need them.
- Prolific — The transparency benchmark: a self-serve participant platform with fully published pricing and an enforced participant pay floor.
- Kili Technology — A Paris-built enterprise platform for annotation and LLM evaluation, sold to regulated industries that need data quality controls more than raw labeling volume.
- SuperAnnotate — A multimodal annotation platform that has grown an expert-services arm (SME) hiring domain specialists for AI training and evaluation work.
- Encord — A data infrastructure platform — Annotate, Index, Active — built for physical AI: robotics, autonomous vehicles and sensor-heavy multimodal data.
- CVAT — The open-source annotation workhorse: born at Intel, now run by CVAT.ai as a free self-hosted Community edition plus paid cloud and enterprise tiers with published prices.
- Label Studio (HumanSignal) — The most popular open-source labeling tool — multimodal by design — with HumanSignal selling cloud and enterprise editions and optional expert data services.
- BasicAI — A platform-and-services company built around multi-sensor data — LiDAR fusion annotation for automotive and robotics is its home ground.
- V7 — The London platform company behind Darwin — AI-assisted visual annotation with a low-cost self-serve entry — now also selling document-workflow agents as V7 Go.
- Dataloop — An Israeli data-pipeline and annotation platform with a strong autonomous-vehicle story — acquired by Dell Technologies in December 2025.
Frequently asked questions
What does Lightly AI do?
An ETH Zurich spin-off built on one idea: label less, but label the right data — curation and self-supervised pretraining first, annotation second.
How much does Lightly AI cost?
Lightly AI does not publish pricing (not published — platform and services quoted; open-source tools free). Any figure quoted by third parties is an estimate; budget discovery requires engaging the vendor directly.
Who is Lightly AI best for?
Vision teams sitting on far more unlabeled data than labeling budget, who want curation to decide what humans should label.
Who should not use Lightly AI?
Text, audio or LLM-feedback programs — Lightly is a computer-vision company.
Sources
- Lightly company profile — CB Insights (accessed Nov 2026)
- Best Data Annotation Companies in 2026 — Lightly blog (vendor-published, read critically)
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.
Profile last updated: November 8, 2026 · How we write profiles · Corrections: [email protected]