Analytics > Member Performance now supports selecting up to 1000 members per view, up from 50. The member list also follows your workspace or project selection, so you're choosing from the right people instead of your entire org.
Analytics KPI data is now available through the API and the label-studio-sdk.
Annotator, reviewer, and other user filters in the Data Manager now show people with access to your selected project instead of your entire organization. A filter also keeps showing people relevant to its specific column, including past contributors who later lost project access, so historical filters still resolve correctly.
Override your keyboard shortcuts for an individual project without changing your account-wide defaults. Edit overrides for any project you have access to from My Account > Hotkeys.
Data Manager "Annotations" filters now support additional child filters. Filter by annotator(s), ground truth status, review outcome, reviewer, comment author, and select multiple values per filter at once.
Right-click any row in the Data Manager, the Organization Members table, or a project's Members page to get the same actions previously available only from the overflow menu, plus your browser's standard right-click options like opening a link in a new tab.
Data Manager filters for Annotators, Reviewers, Comment Authors, Updated By, and Skipped By now accept more than one value at a time. Build a single view across several teammates instead of switching filters back and forth or exporting multiple views to compare them.
Data Manager annotation and prediction result filters now cover more granular dimension-level data, including results produced by Interfaces with per-row AND/OR logic for combining conditions. Narrow down to exactly the results you need to review or export.
As your project progresses, you may find that you need to add metadata columns or update fields along the way. Now, you can more easily add new data columns update values through the UI.
Additional fields (e.g. batch, priority, source, slice_id) can be used as filterable/sortable columns to more easily create the most useful task views.
Instead of having to scroll through different annotators on top tabs, you can now display annotations vertically on the left-hand side.
This is helpful when one task may have multiple annotations, or when you want to search for a specific annotator or label.
In a task, open up the Settings button on the bottom, and choose "Display annotations in vertical panel".
Select multiple tasks and bulk Accept or Reject annotations from All Annotators or a certain Annotator.
This enables you to process reviews faster, especially from trusted annotators or high-agreement tasks.
SDK 2.1.0 is now available on PyPI. This release focuses on stronger typing across the SDK, along with expanded response objects and new filtering options.
pause, members, model_params, and workspace, now use proper object, boolean, integer, and number types in requests and responses.Role9e7Enum is now OrganizationRoleEnum), and all id path parameters are properly typed as integers.For full details, see the release notes on GitHub.
There is a new VideoVector tag that brings Segment Anything 2 (SAM 2) video segmentation and tracking into your annotation workflows.
With SAM 2 connected, you can segment an object once using points or bounding boxes, and the model will track it across frames as the video plays, even when the object moves in and out of frame. This removes the need to segment objects frame by frame.
To get started, you will need an ML backend (an open-source example is available) and GPU-backed hosting for your model. Then, in Label Studio:
For more information, see the blog post or watch the 2-minute video tutorial.
There are now two new dashboards available for projects: Throughput for tasks progress and Data Quality for agreement and label distribution. The Members dashboard has been updated with additional dimension-level metrics and reviewer work, and it is now also available under the Dashboards tab at the project-level.
Learn more in the docs
We're introducing Interfaces, an agentic builder for creating highly customized, multi-modal labeling UIs that you can reuse across projects.
Instead of selecting a template and tuning XML tags, use natural language to describe the labeling experience, and Label Studio generates a fully interactive interface that you can preview, refine, save, and share. With Interfaces, you can:
Review workflows, Agreements, Analytics, exports, and Prompts LLM-as-a-judge and pre-labeling workflows are supported with Interfaces.
Interfaces are available to users in the Owner, Admin, and Manager roles, and can be enabled or disabled for your organization from Organization > Settings > Interfaces.
For more information, see Interfaces.
You can now create service accounts for your organization.
A service account is granted programmatic access to Label Studio via our API, but cannot login or interact with the Label Studio UI.
By default, each organization can create one service account. This service account does not count towards your license's seat count.
You can create and manage service accounts from Organization > Settings > Service Accounts.
For more information, see Service accounts.
You can now filter by is any of and is none of.
These filters accept value lists and can be used with the following fields:
Made several security enhancements by updating key dependencies.
Made performance improvements around image loading.
There is a new section under Project > Settings > Review that you can use to configure how many tasks should be sampled and sent to reviewers.
For more information, see Review Sampling.
You can now publish projects by clicking the project state badge or from Settings > General.
You can now bulk assign members to projects or workspaces from the Organization > Members table.
If the settings for the Agreement columns in the Data Manager have been modified, you will now see a dot indicator icon.
You can now apply a custom color to project and workspace cards:
As we continue to improve dashboards and analytics, the location of the project-level Members dashboard has changed. You can now find it alongside the project dashboard.
The project dashboard is now labeled Throughput.
BrushLabels. TimeSeriesLabels.
The Organization page navigation has been updated to consolidate settings locations.
You will now see a per_control_tag_agreements field in task exports, which maps each active control tag name to its agreement percent (0–100) for the task.
Publishing controls for projects have been moved to the breadcrumbs menu.
To publish a project, click the pause icon ▶️ in the breadcrumbs menu.
When creating new projects, the option to set up cloud storage is more prominently visible with a link to the cloud storage setup workflow.
Improved performance for filtering and sorting per-control-tag agreement.
Improved performance when filtering for compatible projects when creating a Prompt.
If you are using a tag that supports the perItem="true" parameter (for example, Choices, Rating, Taxonomy, Number, TextArea), you can now apply those annotations to individual PDF pages.
You can now view thumbnails for each page. These thumbnails will also provide information about any page-level annotations.
Administrators can now configure ReactCode permissions from Organization > Usage & License.
From here you can configure whether to enable ReactCode or whether to restrict it so that it is only allowed in external app mode.
You can also configure allowed origins and which browser features that iframes can access.
VideoRectangle and Labels tags.
We have made multiple changes and improvements to agreements.
You can now select between Pairwise Averaging and Consensus methodology when calculating agreement.
Previously, all agreement calculations used Pairwise methodology.
You can learn more about each type of methodology in our documentation: Methodology.
Under Settings > Quality > Agreement Columns, you can now select which metric you want to use for each individual control tag.
Before:
Previously, if you had multiple control tags in your labeling interface, you were limited to either the Exact Match metric or the Basic Matching metric, which would apply the default metric to each control tag.
After:
Now you can select a built-in metric for each control tag.
You can also now decide which classes within each tag should be included when calculating agreement.
Under Settings > Quality > Agreement Columns, control tags that have classes (e.g. Choices and Labels) lists each class next to a checkbox.
You can deselect the checkbox to remove that class from the agreement calculation for that control tag. This also removes it from the overall agreement calculation for the task.
As mentioned in an earlier changelog, how weights are configured for agreement has changed.
When you customize the weight for a control tag, you are deciding how much it should factor into the overall task agreement calculation. This lets you de-emphasize tags that should carry less weight when deciding task agreement.
Before:
Previously, you could configure weight by control tag and classes within the control tag.
After:
Now, you can only configure weight by control tag. (However, as mentioned above, you can remove classes from agreement calculation at the tag level instead.)
The Data Manager now has columns for each control tag. This has replaced the Agreement (Selected) column.
You can use the agreement columns to select which annotators and/or models to use in the agreement calculation.
For more information, see Agreement columns.
As mentioned in a previous changelog, the following agreement metrics for rectangles, brushes, and polygons have been removed:
Any existing projects that use the agreement metrics listed above automatically updated to use standard IoU calculation.
Manage Quality at Scale with Super-granular Agreement
Video - Understanding Agreement Metrics with Thresholds
Video - Understanding Agreement | Consensus vs. Pairwise
You can now add metadata tags to organization members. This can be useful if you need to keep track of member location, language, or other attributes.
You will now see:
For more information, see Tag members.
The SSO/SCIM settings are now located under Organization > Settings > SSO & SCIM.
These settings have also been improved to use a more guided setup process with automatic XML/URL metadata field mapping.
There is a new Analytics > Label Distribution dashboard that provides insight into how labels are distributed within a project.
For more information, see Label Distribution dashboard.
When viewing the Overview and Member Performance dashboards, you can now save your filters as a view.
For more information, see Save views.
HyperText tag with valueType=url did not presign URLs within loaded HTML content.
You can now double-click to quickly resize panels in the template builder.
A new custom permission is available to control who can access the Members dashboard for a project: Access Project Members Dashboard
The project list view now includes the project ID and the project description has been moved below the project title.
Added staggered requests to improve performance when loading a large number of images from source storage.
<Markdown> text would overflow the view port when configuring the labeling interface.HyperText links not fully resolving URIs. Data Manager filters have the following enhancements:
Improved performance when clicking between different tasks in the Data Manager to open the Quick View interface.
To support upcoming enhancements to agreement calculations, we will be removing the Agreement (Selected) column and replacing it by adding new functionality to the Agreement column.
While all the functionality offered in Agreement (Selected) will still be available going forward, any filters based on this column will be removed and will need to be recreated.
There is a new Analytics > Overview page that provides insight into metrics and progress across multiple projects and workspaces.
For more information, see Projects overview dashboard.
There are two new tags for image annotation: Vector and VectorLabels.
You can use these tags for point-based vector annotation (polylines, polygons, skeletons).
minPlaybackSpeed parameter could be configured with a value greater than the defaultPlaybackSpeed parameter on the Video tag.