Labelbox, for years one of the first names on any shortlist for a data annotation platform, now opens its homepage with AI agents and reinforcement learning environments for frontier labs. Its labeling editors still exist, filed under a product called Horizon, but its own deprecations page shows the prompt and response editor was switched off on 30 June 2026. Meanwhile the open source Label Studio repository has passed 28,000 stars on GitHub. The market has quietly split into platforms chasing frontier labs and platforms that still want to be the everyday tool on an ML team’s desk.
This guide compares ten options for teams that label more than pictures: text, documents, audio, LLM outputs and mixed data. If your work is mostly images and video, our companion list of computer vision annotation tools goes deeper on that. Every platform below was checked on its official site in October 2026. First on the list is not software at all, but the people who can run any of it for you.
The data annotation platforms at a glance
| # | Option | What it is | Data types | Best for |
|---|---|---|---|---|
| 1 | Zilo AI | Managed annotation teams that work inside your platform | Image, video, audio, text, LLM feedback | Teams that have a tool but not the people |
| 2 | Label Studio | Open source platform, plus a paid Enterprise edition | Almost everything, including LLM evaluation | Teams that want one flexible tool they can self-host |
| 3 | Labelbox | Commercial platform now aimed at frontier AI labs | Multimodal, preference and rubric data | Large labs buying tooling and expert labelers together |
| 4 | SuperAnnotate | Commercial multimodal data platform | Image, video, audio, LiDAR, RLHF, SFT, RL environments | Teams mixing classic labeling with LLM post-training |
| 5 | Encord | Commercial data platform with a physical AI focus | Video, LiDAR, audio, text, documents, DICOM | Robotics, autonomy and medical imaging teams |
| 6 | Kili Technology | Commercial annotation and evaluation platform | Image, video, text, OCR, geospatial, audio | Document-heavy and evaluation-heavy enterprise work |
| 7 | CVAT | Open source, cloud and enterprise editions | Image, video, 3D point clouds, audio | Vision-first teams adding audio |
| 8 | Prodigy | Downloadable, scriptable annotation tool | Text first, plus images, audio and video | Python developers building NLP models |
| 9 | INCEpTION | Open source semantic text annotation | Text with knowledge-base linking | Research corpora and entity linking |
| 10 | doccano | Open source text annotation tool | Text classification, sequence labeling, seq2seq | Small teams starting an NLP dataset |
1. Zilo AI
Zilo AI is a managed annotation service, not a data annotation platform, and that is exactly why it sits first here. Most teams that buy a platform discover the harder problem a month later: someone has to sit in the tool and label tens of thousands of items to a written guideline, and someone else has to review them. Zilo supplies those people. Its Bangalore teams work inside the platform you already use, and its own service pages name CVAT, Label Studio, Labelbox, SuperAnnotate, Encord, V7, Prodigy, doccano and INCEpTION among the tools they work in.
The work covers image, video, audio and text annotation, transcription, RLHF and AI evaluation. Zilo says it has over 1,600 trained annotation and ASR experts and has annotated more than 10 million data points for industries from retail to BFSI to healthcare. Engagements run as managed projects, pilot batches or dedicated teams.
Who it suits: teams that have picked a platform, or are about to, and need a trained workforce and a review layer rather than more software. What to check: run a pilot batch on your own guideline in your own tool, and agree how disagreements and edge cases get escalated before volume starts.
2. Label Studio
Label Studio is the open source platform with the largest GitHub following on this list. It is maintained by HumanSignal, released under the Apache 2.0 license, and installs with a single pip command. The project page lists computer vision, documents and NLP, audio and speech, time series and multimodal data, and it has moved firmly into LLM work: human preference collection for RLHF, side-by-side model comparison, custom benchmarks and grading of retrieval answers. HumanSignal says more than a million AI practitioners use, build and support it.
The paid Enterprise edition adds AI-assisted labeling, review workflows, workforce management and SOC 2 and HIPAA-oriented security, deployed in HumanSignal’s cloud or on your own servers. HumanSignal also sells annotation services if you want the vendor to staff the work.
Who it suits: teams that want one configurable tool across several data types and are comfortable writing labeling templates. What to check: which features you need sit in the free edition and which are Enterprise only, especially reviewer workflows and role permissions.
3. Labelbox
Labelbox remains a serious annotation business, but its centre of gravity has moved. The homepage now sells three things: Recursion, a managed agents product; Horizon, which covers RL environments, evaluations and the labeling platform; and Alignerr, an expert labeling network Labelbox says has more than 2.6 million skill-tested experts across 200 domains. The labeling side still has editors, quality tools, a data catalog and labeling services, documented under Horizon.
Its own deprecation log is worth reading before you commit. Beyond the prompt and response editor retired in June 2026, the DICOM editor was sunset in November 2024 and the custom editor in June 2024.
Who it suits: well-funded AI teams that want software and expert labelers from one vendor, particularly for preference data and rubric grading. What to check: whether the specific editor you need is on the current roadmap, and whether a standard annotation project still gets attention from a vendor now built around frontier labs.
4. SuperAnnotate
SuperAnnotate describes itself as AI data infrastructure and pitches one editor for every modality: RLHF preference data, supervised fine-tuning, image, video, audio, LiDAR and custom RL environments with verifiers and rubric scoring. Your own models can pre-label items and judge rubrics, and orchestration rules route each item between models and human experts. The company says it is ranked the top data annotation platform on G2.
Who it suits: teams running classic labeling and LLM post-training side by side who want one workflow system for both. What to check: how much of the orchestration you will actually configure yourself, and what the pricing looks like once model-assisted steps run at volume.
5. Encord
Encord calls itself the data layer for physical AI and enterprise teams. Its annotation tooling handles native video, LiDAR, audio, text and sensor fusion in the same workflow, with label lineage and quality controls built in. DICOM and NIfTI medical formats and geospatial data are available as add-ons. Beyond software, Encord now offers managed annotators and data collection from its own teleoperation facilities. Plans run from a Starter tier for prototypes to Enterprise with SSO and VPC or on-premise deployment.
Who it suits: robotics, autonomy and medical imaging teams whose data arrives as video and sensor streams. What to check: which data types are included in your plan and which are paid add-ons.
6. Kili Technology
Kili Technology now calls itself an AI evaluation, post-training and data annotation platform, in that order. It covers video, image, NLP and LLM work, OCR and document processing, geospatial imagery and audio. Its emphasis is on expert review: built-in quality checks, an audit trail of decisions, and deployment on Kili’s cloud, your own storage or your own Kubernetes cluster. For selected projects, Kili’s partner network sources and trains annotators and Kili manages quality.
Who it suits: enterprises labeling documents, forms and text for extraction, or running structured human evaluation of models. What to check: the deployment option your security team will accept, and the pricing, which is best confirmed directly with sales.
7. CVAT
CVAT is best known as a computer vision tool, and our computer vision annotation tools guide covers it in depth. It earns a place on a general list because it now labels audio too: its homepage lists speech transcription into time-aligned text and detection of speech and sound events, next to images, video and 3D point clouds. It comes as a free open source Community edition, a hosted CVAT Online with free, solo and team plans, and a self-hosted Enterprise edition with SSO.
Who it suits: vision-led teams that have started collecting audio and want to keep one tool. What to check: whether CVAT’s audio features cover your transcription rules, since they are newer than its vision tooling.
8. Prodigy
Prodigy takes the opposite approach to the cloud platforms. You buy a lifetime license, install it like a Python library and run it on your own machine, so no data leaves your servers. Annotation flows are written as Python “recipes”, and the case studies on its site pair it with the spaCy NLP library. It is strongest on text, such as entity recognition and classification, and also handles image, audio and video tasks and LLM prompt evaluation. Its maker says more than 10,000 developers and researchers use it.
Who it suits: small, technical NLP teams that want speed and privacy more than a management dashboard. What to check: whether you need multi-annotator management and reporting at scale, which is where hosted platforms do more of the work for you.
9. INCEpTION
INCEpTION is an open source platform for multi-layer text annotation, built for corpora where every mention should link to an entry in a knowledge base. A recommender learns from your annotations as you go and suggests the repetitive ones. It is actively developed: version 42.1 was released on 13 October 2026, and the project is moving to the Eclipse Foundation.
Who it suits: research groups, digital humanities projects and teams building entity-linked text datasets. What to check: that you have someone to host and administer it, because it is a self-run application rather than a service.
10. doccano
doccano is the simplest tool on the list, which is its appeal. It is an open source text annotation tool for text classification, sequence labeling and sequence-to-sequence tasks, so it covers sentiment, named entities and summaries. You can run it through pip or Docker, and it has a REST API and collaborative annotation. The repository has more than 10,000 GitHub stars, and its latest release came out in January 2026.
Who it suits: small teams and students starting a first NLP dataset. What to check: the release pace, which is slower than the commercial platforms, and whether you will outgrow its review features.
Why some familiar names are missing
Earlier versions of this page listed Appen, CloudFactory, Scale AI and Fiverr. Appen, CloudFactory and Scale AI are best understood as service companies rather than software you run. Appen and CloudFactory now appear on our list of the best data annotation companies, and Scale AI on our list of LLM and RLHF data companies. Fiverr is a freelance marketplace, not annotation software. We also left out Amazon SageMaker Ground Truth, which AWS has closed to new customers, according to its documentation.
How to choose a data annotation platform
- You need one tool for many data types and want to self-host: Label Studio, or CVAT if images and video dominate.
- You are doing LLM post-training as well as classic labeling: SuperAnnotate, Label Studio Enterprise or Kili.
- Your data comes from robots, vehicles or scanners: Encord.
- You are a small NLP team that writes Python: Prodigy, or doccano if the budget is zero.
- You are building a research corpus with entity linking: INCEpTION.
- You have the tool but not the people: Zilo AI, working inside whichever platform you choose.
Whatever you shortlist, label the same 200 items in two platforms before you sign anything. Time how long your reviewers take, export the results into your training format, and check that the export matches what your model code expects. If you want a team to run that pilot for you, talk to Zilo AI.
Frequently asked questions
What is a data annotation platform?
It is software where people, often helped by models, attach labels to raw data such as images, text, audio or model outputs, and where managers set guidelines, assign work, review quality and export the finished dataset.
What is the best free data annotation tool?
For most teams it is Label Studio’s open source edition, because it covers the most data types. CVAT’s Community edition is the strongest free choice for images and video, and doccano is the simplest free option for text.
What is the difference between annotation software and an annotation service?
Annotation software is the tool. An annotation service supplies the trained people who use it, plus the review process. Many teams need both, which is why services such as Zilo AI work inside platforms such as Label Studio, CVAT or SuperAnnotate.
Can I still use Amazon SageMaker Ground Truth?
Only if you were already a customer. AWS has closed Ground Truth to new customers, and Amazon Mechanical Turk, which supplied its public workforce, closed on 30 September 2026.
Which data annotation tools support LLM evaluation and RLHF?
Label Studio, SuperAnnotate, Kili, Labelbox and Prodigy all support preference ranking or model output evaluation in some form. Check whether the feature is in the edition you plan to buy.