CVAT vs Label Studio: Open-Source Annotation Compared (2026)
By Daniel Clarke
CVAT and Label Studio are the two default open-source annotation tools, and most comparisons of them are published by one of the two companies. This one isn't. We sell no labeling platform and take no position on which you pick.
Here's the short version: CVAT is the better computer vision tool. Label Studio is the better everything-else tool. If your data is images, video, or 3D point clouds, use CVAT. If it includes text, audio, time series, or you're evaluating LLMs, use Label Studio. Most teams can stop reading there.
If you're weighing a closer call, the rest of this page covers the three things that actually decide it: what each tool gates behind a paid tier, how much self-hosting really costs you, and whether both projects are healthy.
GitHub data pulled 2026-07-17. Pricing and tier limits last checked 2026-07-17.
Data type support: the decision most teams need
| Data type | CVAT | Label Studio |
|---|---|---|
| Images | Yes | Yes |
| Video | Yes (strong) | Yes |
| 3D / LiDAR / point cloud | Yes | No |
| Text / NLP | No | Yes |
| Audio | No | Yes |
| Time series | No | Yes |
| Documents (PDF / OCR) | No | Yes |
This table is the whole comparison for most people. CVAT supports images, video, and 3D data — and nothing else. Label Studio covers six data types but has no point cloud annotation.
If you need 3D/LiDAR and text, neither tool does both. That's a real gap, and it's the honest reason some teams end up on Labelbox or Encord instead.
What's actually gated behind the paid tier
This is the most important section on this page, and the one no vendor publishes clearly. "Open source" does not mean the same thing for these two tools.
| Capability | CVAT (self-hosted) | Label Studio (Community) |
|---|---|---|
| Core annotation | Included | Included |
| API / SDK | Included | Included |
| Import / export | Included | Included |
| Role-based access control | Included | Enterprise only |
| SSO | Included | Enterprise only |
| Review / QA workflows | Included | Enterprise only |
| Reviewer assignment | Included | Enterprise only |
| Advanced analytics | Included | Enterprise only |
| SOC 2 compliance | Enterprise offering | Enterprise only |
| Licence | MIT | Apache-2.0 |
The asymmetry matters. Self-hosted CVAT gives you the full product for free. Label Studio's Community Edition is genuinely powerful, but role-based access control, SSO, reviewer assignment, and QA workflows are Enterprise features.
The practical consequence: if you have a team of annotators who need to be managed, reviewed, and permissioned, Label Studio Community will not do that. A solo researcher or a small trusted team won't notice. A 20-annotator operation with a review stage will hit this wall fast, and the answer is a sales call.
If your reason for choosing open source is "we don't want to talk to sales," CVAT holds that promise further than Label Studio does.
Project health: real numbers, not vibes
Nobody publishes this, so here it is. Pulled from the GitHub API on 2026-07-17:
| Signal | CVAT (cvat-ai/cvat) |
Label Studio (HumanSignal/label-studio) |
|---|---|---|
| Stars | 16,317 | 27,855 |
| Forks | 3,764 | 3,624 |
| Open issues | 601 | 900 |
| First commit | June 2018 | June 2019 |
| Latest commit | 2026-07-16 | 2026-07-17 |
| Commits in last 90 days | 100+ | 100+ |
| Latest stable release | v2.70.0 (2026-07-14) | 1.23.0 (2026-03-13) |
| Release cadence | ~every 2–4 weeks | ~quarterly (+ rolling nightlies) |
Both projects are healthy. Both exceeded 100 commits in the last 90 days (that's the API's page limit, so the true figure is higher for both), and both had commits within a day of this being written.
The difference is release philosophy, not activity. CVAT ships small versioned releases roughly every two to four weeks — v2.66 through v2.70 all landed between late May and mid-July 2026. Label Studio ships stable releases roughly quarterly (1.20 in July 2025, 1.21 in September, 1.22 in December, 1.23 in March 2026) with a continuously updated nightly tag in between.
Don't read Label Studio's four-month gap since 1.23.0 as abandonment — the repo is committed to daily. But if you self-host and want fixes in tagged, tested releases rather than running nightlies, CVAT gets them to you faster. If you prefer fewer, larger, more stable upgrades, Label Studio's cadence is a feature.
Label Studio has ~70% more stars and a bigger community (a 20K+ member Slack, 1M+ users). CVAT has the OpenCV Foundation behind it and originated at Intel.
Self-hosting: what it actually costs you
Both are free to self-host. Neither is free to run.
Label Studio is the easier start: it installs via pip, brew, or Docker, and you can be annotating in minutes on a laptop.
CVAT is Docker Compose from the outset and assumes more infrastructure familiarity.
That gap closes fast in production. Once you add cloud storage, backups, auth, upgrades, and enough RAM, both are a standing ops commitment. Two caveats apply to both, and they're the most common self-hosting complaints in our reviews:
- Performance degrades on large datasets. CVAT struggles with very large video files and projects with thousands of images. Label Studio has the same reported issue at scale. Neither is a hosting bug you can configure away — budget real hardware.
- The learning curve is real, differently. CVAT's complexity is its interface. Label Studio's is its XML template system — powerful enough to build almost any labeling UI, and steep enough that non-technical annotators struggle at first.
The honest framing: self-hosting trades a vendor invoice for engineering hours. For a team with DevOps capacity that's usually a good trade. Without it, the managed tiers below are cheaper than they look.
Pricing, if you'd rather not self-host
| CVAT | Label Studio | |
|---|---|---|
| Self-hosted | Free (MIT) | Free (Apache-2.0) |
| Free cloud tier | 1 project, 3 tasks, 1GB, 100 AI calls/mo | — |
| Entry paid plan | $23–33/mo (Solo) | $50/mo (Starter Cloud) |
| Team plan | $23–33/user/mo | Custom |
| Enterprise | ~$12,000/year | Custom pricing |
CVAT is the more transparent of the two: it publishes a team price and even an approximate enterprise figure, which is rare in this category. Its free cloud tier is very limited — 1 project and 3 tasks is an evaluation sandbox, not a workspace.
Label Studio publishes only the $50/mo Starter Cloud entry point; Enterprise is a sales conversation.
AI-assisted labeling
CVAT integrates SAM 2/3 for interactive segmentation and offers auto-annotation with built-in or custom models, claiming up to 10x faster labeling. Label Studio leans toward LLM and GenAI workflows: custom benchmarks, RLHF, RAG evaluation, and response moderation.
They're not competing here so much as serving different jobs. CVAT accelerates drawing boxes and masks. Label Studio helps you evaluate model outputs.
Choose CVAT if
- Your data is images, video, or 3D/LiDAR — and only those
- You want the full product self-hosted with no feature gating
- You need CV export formats (CVAT supports 19, including YOLO, COCO, PASCAL VOC, KITTI)
- You want frequent tagged releases
- You want a published team and enterprise price
Read our full CVAT review.
Choose Label Studio if
- Your data includes text, audio, time series, or PDFs
- You're building LLM evaluation, RLHF, or RAG validation workflows
- You need a custom labeling interface that doesn't exist off the shelf
- You want the fastest possible local start (
pip install) - You want the larger community
Read our full Label Studio review.
Can you run both?
Yes, and plenty of teams do — CVAT for the vision datasets, Label Studio for text and evaluation. They're free, so the only real cost is maintaining two deployments and two sets of annotator training.
Worth it if both workloads are substantial. If one is marginal, a single multimodal platform like Labelbox (free tier: 30 users, 50 projects) is usually less overhead than two self-hosted tools.
Frequently asked questions
Is CVAT or Label Studio better? Neither, in the abstract. CVAT is better for computer vision, especially 3D/LiDAR. Label Studio is better for multimodal data and LLM evaluation. The data you have decides it.
Are both really free? Both are free to self-host under permissive licences (CVAT is MIT, Label Studio is Apache-2.0). The difference is what "free" includes: CVAT self-hosted is the full product, while Label Studio Community excludes RBAC, SSO, and review workflows.
Which is easier to set up? Label Studio, at first — pip, brew, or Docker versus CVAT's Docker Compose. In production the gap narrows and both need real ops attention.
Does Label Studio support LiDAR or point clouds? No. That's CVAT's clearest advantage. For 3D beyond CVAT, look at Encord or Supervisely.
Does CVAT support text or audio annotation? No. CVAT is computer vision only. Use Label Studio, or a multimodal platform like Labelbox.
Which is better for a large annotation team? CVAT self-hosted, if you're avoiding paid tiers — you get permissions and review workflows at no cost. Label Studio gates those behind Enterprise, so a large managed team means paying or building around it.
Are these projects actively maintained? Yes, both. Each had 100+ commits in the 90 days before 2026-07-17 and commits within a day of this writing. CVAT tags releases every few weeks; Label Studio ships quarterly stable releases plus nightlies.
For the wider field, see the best open-source data labeling tools.