There is no single best data labelling company, because the market sells several different things under the same name: crowds, managed workforces, expert networks for frontier models, software for your own team, and licensed datasets. This guide groups 22 providers by the job they do best, so you can shortlist three and run a pilot instead of reading twenty sales pages.
The short list
Disclosure. Dataset Finder publishes this guide and is one of the options in it. We have kept to facts each company states or that were reported in the press, and linked our detailed comparison pages so you can check the trade-offs yourself.
Providers are grouped by use case, not ranked: a frontier lab, a start-up with 5,000 images and a hospital with DICOM scans should not end up with the same vendor.
| Option | Type | Pricing | Best for |
|---|---|---|---|
| Dataset Finder (ours) | Dataset search + self-serve custom data | Free plan; Startup €9.90/month; Pro €29.90/month. Custom data paid in credits, with a public cost estimator. | Teams that want to check what already exists, then order the gaps self-serve |
| Toloka | Managed service + self-serve crowd | Not published; self-service priced per project | RL environments and evaluation, or small crowd tasks |
| Label Studio (HumanSignal) | Open-source platform + services | Free open source; $99/month Starter Cloud; Enterprise on request | A flexible tool your team can self-host |
| CVAT | Open-source platform + services | Free open source and online plan; paid from $23/month | Budget-conscious computer-vision teams |
| Sama | Managed service | Not published; quote-based | High-volume image, video and LiDAR annotation |
| iMerit | Managed service + platform | Not published | Medical imaging and mobility at enterprise scale |
| Appen | Managed service + data catalogue | Not published | Multilingual, speech and search-relevance programmes |
| TELUS Digital | Managed service | Not published; sales-led | High-volume, multilingual and automotive programmes |
| CloudFactory | Managed workforce + platform | Not published | Ongoing human review for AI already in production |
| Scale AI | Managed service | Not published; sales-led | Frontier labs and government programmes at very high volume |
| Surge AI | Managed service | Not published; sales-led | Frontier labs buying high-skill RLHF at scale |
| Mercor | Expert network | Not published; sales-led | AI labs that need credentialed domain experts fast |
| Turing | Expert network + managed service | Not published; sales-led | Code and reasoning post-training data |
| Invisible Technologies | Managed service + expert network | Not published; sales-led | A fully managed expert-data team |
| Snorkel AI | Managed service (expert data) | Not published; sales-led | Expert post-training data and evaluations |
| Labelbox | Platform + expert network | Free tier; Starter and Enterprise; hourly rates for labelling services | Your own team in a platform, or expert preference data |
| SuperAnnotate | Platform + services | Starter, Pro and Enterprise; prices not published | Running your own annotation and evaluation pipelines |
| Encord | Platform + Data-as-a-Service | Starter, Team and Enterprise; prices not published | Robotics, AV and medical-imaging teams with their own data |
| Kili Technology | Platform | Free trial; Grow and Enterprise on request | Regulated European teams labelling with their own experts |
| V7 Darwin | Platform + services | Custom annual quote | Medical imaging and video with AI assist |
| Defined.ai | Data marketplace | Not published; samples and quotes on request | Licensed speech and multilingual data |
| Shaip | Data marketplace + services | Not published | Licensed medical and speech corpora |
Five kinds of labelling company
Many companies now span two or three models: platforms sell labelling services, managed services build platforms, and expert networks sell off-the-shelf datasets. The grouping below follows what each is best known for.
| Model | Who does the work | How you pay | Fits |
|---|---|---|---|
| Self-serve and crowd | Managed annotators or a crowd, ordered online | Credits, per task or per project, often no minimum | Small and mid-size projects, fast starts |
| Managed workforce | Employees or contractors in delivery centres | Quote per project or per hour | High volumes of your own data |
| Expert network | Vetted domain experts, often paid hourly | Quote; per expert hour | RLHF, SFT and evaluations for frontier models |
| Platform | Your own team, optionally the vendor’s annotators | Seats and usage, sometimes a free tier | Expert knowledge in-house, data that must stay put |
| Marketplace | Already collected and labelled | Licence per dataset | Speech, medical and multilingual corpora |
Self-serve and small-to-mid projects
For projects from a few hundred to a few hundred thousand items, where speed and a visible price matter more than a bespoke programme.
Dataset Finder (ours)
Dataset search + self-serve custom dataSearch tens of thousands of curated datasets in plain language, then order the data that does not exist yet from Innovatiana’s managed annotators in the same app, paid in credits.
- Each dataset card adds automated analysis on 30+ criteria, including licence, PII, bias and fine-tuning readiness.
- Custom requests are labelled by trained annotators and domain specialists under a dedicated Data Labeling Manager, not an open crowd.
- A test batch usually comes back within about 48 hours; large projects move to an Enterprise quote.
- Launched in August 2026, so the product is new; the annotation work is done by Innovatiana, the data-labelling company behind it.
Toloka
Managed service + self-serve crowdExpert data, RL environments and evaluation for AI labs, plus a self-service crowd platform with no minimums.
- Raised $72M in a round led by Bezos Expeditions in May 2025.
- Contributors in 100+ countries and experts in 90+ domains, Toloka says.
Label Studio (HumanSignal)
Open-source platform + servicesThe widely used open-source labelling tool for text, audio, images, video and time series, with hosted and enterprise editions.
- The Community Edition is open source under Apache 2.0.
- Starter Cloud costs $99 a month for up to 12 users; Enterprise is priced on request.
- Launched HumanSignal Services for data creation in November 2025, alongside its acquisition of Erud AI.
CVAT
Open-source platform + servicesThe Computer Vision Annotation Tool, started at Intel and spun out in 2022, for image, video and 3D annotation.
- Open source under the MIT licence, with a free online plan.
- Paid online plans from $23 a month billed yearly; self-hosted Enterprise priced by sales.
Managed workforces for high-volume annotation
The classic outsourcing model: a vendor-managed team labels your images, video, LiDAR, audio or text at volume, on its platform or yours.
Sama
Managed serviceManaged annotation and model validation on its own platform, with a full-time workforce in secure delivery centres.
- Sama says it was one of the first AI companies certified as a B Corp, and has been recertified.
- Depth in automotive, retail and 3D sensor annotation; Sama reports 99% first-batch acceptance.
iMerit
Managed service + platformManaged annotation and GenAI data on its Ango Hub platform, with strong medical-imaging and autonomous-mobility work.
- EXL completed its acquisition of iMerit on 3 August 2026.
- Launched iMerit Scholars, an expert network for GenAI training, in July 2025.
Appen
Managed service + data catalogueCustom collection, annotation and evaluation, an off-the-shelf dataset catalogue and the ADAP platform.
- A crowd of over 1 million contractors in 170+ countries, Appen says, covering 500+ locales.
- Listed on the Australian Securities Exchange; revenue and other income rose 17% in the half year to June 2026.
TELUS Digital
Managed serviceThe AI data arm of TELUS Digital (formerly TELUS International): data collection, multimodal and 3D annotation, RLHF and red-teaming.
- TELUS completed its buy-out of TELUS Digital on 31 October 2025, taking it private.
- Says it has an AI community of over 1 million contributors and covers 500+ annotation languages and dialects.
CloudFactory
Managed workforce + platformA managed human-in-the-loop workforce, an AI platform and consulting, since 2010.
- Lists ISO 27001:2022, SOC 2 and a HIPAA attestation.
- Its homepage now leads with an AI platform and consulting rather than labelling alone.
Expert networks for frontier post-training
Preference data, SFT, rubrics, evaluations and RL environments written by vetted professionals, mostly for the largest AI labs. Several of these companies raised large rounds in 2025 and 2026.
Scale AI
Managed servicePost-training and evaluation data for AI labs at very high volume, plus AI applications for enterprises and governments.
- Meta made a minority investment in June 2025 that valued Scale at over $29 billion; founder Alexandr Wang joined Meta.
- Francis deSouza became CEO on 10 August 2026.
- Contributors come through Outlier, an expert network Scale operates (900K+ experts, Outlier says).
Surge AI
Managed servicePremium human data for frontier labs: RLHF, post-training data, RL environments and benchmarks.
- Reuters reported over $1 billion in 2024 revenue, without outside funding.
- Expert-heavy work aimed at the largest AI labs; Surge does not publish the size of its contractor network.
Mercor
Expert networkA marketplace that recruits and vets domain experts such as doctors, lawyers and engineers, and bills AI labs for the training data, evaluations and RL environments they build.
- Raised a $350M Series C at a $10 billion valuation in October 2025.
- Agreed in July 2026 to acquire Deeptune, a start-up building RL training environments for AI agents.
- Its CEO said gross annualised revenue passed $2 billion in June 2026 (a company figure, before contractor payouts).
Turing
Expert network + managed serviceA former remote-developer marketplace that now supplies coding, STEM and knowledge-work experts, datasets and RL environments to AI labs.
- Raised $111M at a $2.2 billion valuation in March 2025.
- Launched Turing Frontier in April 2026, giving AI labs access to identity- and background-verified professionals.
Invisible Technologies
Managed service + expert networkRLHF and SFT data, evaluations and RL environments for AI labs through its Meridial expert network, plus enterprise workflow automation.
- Raised about $100M at a valuation above $2 billion in September 2025.
- Says its Meridial network has 24,000+ vetted experts.
- Agreed in March 2026 to acquire WeCP, a technical assessment platform, to strengthen expert vetting.
Snorkel AI
Managed service (expert data)A Stanford AI Lab spin-out that now sells expert-built training data, RL environments, rubrics and evaluations to AI labs and enterprises.
- Raised a $350M Series E at a $3.5 billion valuation in September 2026.
- Launched Expert Data-as-a-Service in September 2025, with work done by vetted domain experts.
- Also offers off-the-shelf expert datasets with rubrics and difficulty tiers.
Labelling platforms
Software for your own team, with model-assisted labelling, review workflows and, at most vendors, optional annotation services.
Labelbox
Platform + expert networkA labelling platform plus the Alignerr expert network, RL environments (Horizon), robotics data (Terra) and managed agents.
- A free platform tier for up to 30 users and 50 projects.
- Raised a Series E of over $100M, equity plus a credit facility, in October 2026; Labelbox says it is profitable.
SuperAnnotate
Platform + servicesAn annotation and evaluation platform for multimodal data, LLMs and agents, with annotators it can provide.
- Series B extended to $50M in July 2025, with Dell Technologies Capital.
- Evaluation tooling for LLMs and agents, including RLHF ranking and golden-dataset creation.
Encord
Platform + Data-as-a-ServiceA data platform for physical AI: curation, annotation and evaluation of video, LiDAR, DICOM and other multimodal data.
- Raised a $60M Series C in February 2026, bringing total funding to $110M.
- VPC and on-premise deployment options for Enterprise.
Kili Technology
PlatformA self-hostable labelling and evaluation platform for data written by your own experts, built in Paris.
- Self-hosted and on-premise deployment for regulated sectors.
- Customers named on its homepage include SAP, Safran, Michelin and Airbus.
V7 Darwin
Platform + servicesAn AI-assisted annotation platform for images, video and medical imaging, from London-based V7.
- Supports DICOM and NIfTI, microscopy and video, with model-assisted labelling and multi-stage review.
- Also sells labelling services, including radiologists and pathologists for medical imaging.
Licensed datasets
When the data you need may already exist under a commercial licence, a marketplace is faster than collection.
Defined.ai
Data marketplaceLicensable off-the-shelf datasets, especially speech and multilingual, plus custom collection and annotation.
- Lists ISO 42001 alongside ISO 27001 and ISO 27701.
- Reported 65% revenue growth for 2025.
Shaip
Data marketplace + servicesData collection, annotation and licensing, strong in medical and speech data; part of Ubiquity since 2026.
- Shaip’s homepage cites 30M patient notes and 250k hours of medical audio.
- Ubiquity Global Services agreed to acquire it in February 2026.
Questions to ask any labelling company
- Who exactly labels the data? Employees, a crowd, contractors or credentialed experts, and in which countries.
- How is quality measured and reported? Ask for inter-annotator agreement, gold-standard accuracy and review rates on your pilot, and check them yourself with the kappa calculator.
- What does a pilot cost and how fast is it? A test batch before the full run is the cheapest way to catch a misunderstanding.
- Can they work in your tool? Many annotators work in Label Studio, CVAT, Labelbox or Encord, which keeps the audit trail on your side.
- Which certifications and data terms apply? ISO 27001, SOC 2, HIPAA or TISAX if your sector needs them; who owns the labels; NDAs and retention.
- How is the price built? Per hour, per label, per item or per credit. Ask for a quote against written guidelines, and compare it with an estimate.
Sources
Facts were checked on 26 October 2026 against each company’s own site, press releases and filings, and reputable press for corporate events; figures a company reports about itself are marked as its own. Main sources: Dataset Finder, Label Studio (HumanSignal), CVAT, TELUS Digital, Scale AI, Surge AI, Mercor, Turing, Invisible Technologies, Snorkel AI, V7 Darwin. The comparison pages list every source for the vendors they cover.
Questions people ask
What is the best data labelling company in 2027?
It depends on the job. Scale AI, Surge AI and the expert networks lead on frontier post-training data; Sama, iMerit, Appen and TELUS Digital on high-volume managed annotation; Labelbox, SuperAnnotate, Encord and Kili on platforms; and self-serve options such as Dataset Finder, Toloka’s platform, Label Studio and CVAT suit smaller projects and teams that want to start quickly.
How much does data labelling cost?
Most vendors quote per project, per hour or per label, and few publish prices. Cost scales with data type, the number of classes and objects per item, and the expertise needed. The free annotation cost estimator gives a first figure for a custom request.
Is it cheaper to label data in-house?
Often not, once your team’s time is counted: writing guidelines, reviewing and correcting labels come on top of the labelling itself, and recruiting, training and tooling add fixed costs. In-house makes sense when the knowledge cannot be written down or the data cannot leave.
What is the difference between data labelling and data annotation?
In practice the terms are used interchangeably. “Labelling” is sometimes kept for assigning a class to a whole item and “annotation” for richer marks such as boxes, masks, spans or keypoints, but vendors use both for the same services.