Alibaba Cloud PAI iTAG

Alibaba Cloud PAI iTAG is a data-labeling platform for image, text, video, audio, and multimodal data. Its built-in image tasks include classification, object detection, OCR, table recognition, and semantic segmentation. Official templates also cover image-text explanation, conversation ranking, visual question answering, and video OCR; teams can use official or tenant templates or create their own for specific labeling needs. A job brings together a dataset and template with labels, subtasks, checks, labelers, and reviewers. Manual labeling is free, and selected large-model labeling templates are currently free, while OSS storage and data transfer are billed separately. Input data must be stored in Alibaba Cloud OSS, with the workspace and bucket in the same region; imports require an index file rather than a direct folder path. From August 1, 2026, iTAG 1.0 is allowlist-only and in maintenance mode. Existing customers with prior labeling tasks or datasets can continue using current features. New users seeking AI-powered automatic dataset labeling are directed to Multimodal Data Management.

iTAG offers varied labeling templates and a workflow with checks, labelers, and reviewers. Its OSS requirements and maintenance-mode availability are important considerations, particularly for new users.

It suits teams labeling image, text, video, audio, or multimodal data in Alibaba Cloud OSS. New users seeking AI-powered automatic dataset labeling are directed to Multimodal Data Management instead.

Alibaba Cloud PAI iTAG is a managed labeling workspace for image, text, video, audio, and multimodal datasets. It is most useful for organizations already operating in Alibaba Cloud that need configurable annotation jobs and human review. Its range of templates is substantial, but the storage dependency and restricted future availability make it a cautious choice for new deployments.

Overview

iTAG covers common computer-vision work alongside language, speech, video, and multimodal annotation. Image projects can use classification, object detection, OCR, table recognition, and semantic-segmentation tasks. The service is cloud-deployed and managed, with an administrative console, container support, clustering, and API access.

Its strongest fit is a team that already keeps source data in Alibaba Cloud OSS and wants labeling to remain inside that environment. Teams starting with local disks or NAS face an extra data-migration step, while organizations that need a new, long-lived labeling platform should weigh iTAG's maintenance status before committing.

Key features

Template selection is broader than a fixed set of canned forms. Official templates cover image, text, video, audio, and multimodal work; tenant templates support organization-specific reuse; and custom templates can encode a business's own labeling scheme. Multimodal options include image-text explanation, conversation ranking, visual question answering, and video OCR, giving teams a path beyond conventional bounding boxes and class tags.

A labeling job combines a dataset and template, then exposes settings for labels, subtasks, checks, labelers, and reviewers. That arrangement supports a controlled handoff from annotation to checking and review, which is more suitable for managed production workflows than an unstructured single-pass queue. Human review workflows and custom ontologies are supported, and model-assisted labeling is available.

Storage rules are consequential. Input data must reside in Alibaba Cloud OSS; local or NAS files have to be uploaded first. The iTAG workspace and OSS bucket must share a region, and importing a dataset requires an index file rather than a folder path containing raw files. Those requirements can complicate existing pipelines and cross-region layouts.

The OpenAPI can list, create, update, and delete templates and labeling tasks, making it practical to connect job administration to internal tooling. Alibaba Cloud recommends an API-only AccessKey on a RAM user with least-privilege permissions. FAQ material and an Alibaba Cloud AI assistant provide product and documentation support.

Pricing

iTAG manual labeling: free. Manual team labeling does not incur a platform fee, but OSS storage and data transfer are billed separately. This is the natural fit for teams that can supply their own annotators and already budget for Alibaba Cloud storage; it is not an all-in cost-free workflow.

Selected large-model templates, including image-to-text and image-text explanation, are currently free. Alibaba Cloud says users will be notified if pricing changes, so teams building recurring model-assisted volume should allow for a possible future charge. There is no free trial, and no additional paid plan terms are published beyond the free manual-labeling offer.

Platforms

iTAG is available through web and API interfaces, with Linux and Windows support. Its supported operating-system list also includes Anolis OS, Aliyun Linux 2 and 3, Kylin, and Unity Operating System. Deployment is cloud-based and managed rather than an on-premises installation.

Who it's for

Choose iTAG when your data already lives in Alibaba Cloud OSS, your projects span several media types, and you need configurable templates plus reviewer checkpoints. It also suits engineering teams that want to automate template and task administration through OpenAPI.

It is a weaker fit for a first-time labeling rollout, a pipeline centered on local or NAS storage, or a deployment that must remain broadly available to new customers. Since August 1, 2026, iTAG 1.0 has been allowlist-only and in maintenance mode. Existing customers with prior labeling tasks or datasets can continue using current features, while new users seeking AI-powered automatic dataset labeling are directed to Multimodal Data Management.

Pros and cons

• Pro: Covers image, text, video, audio, and multimodal annotation, so one service can handle varied dataset programs.
• Pro: Official, tenant, and custom templates let teams adapt labeling to specialized ontologies and business rules.
• Pro: Labels, subtasks, checks, labelers, and reviewers provide a structured quality workflow.
• Pro: Manual labeling is free at the platform level, and selected large-model templates are currently free.
• Con: OSS is mandatory, with same-region placement and index-file import requirements that add operational friction.
• Con: OSS storage and transfer costs remain payable even when manual labeling costs nothing.
• Con: Allowlist-only maintenance mode limits its appeal for new, long-term implementations.

Alternatives

doccano is the better pick for teams wanting free, MIT-licensed software that runs across web, desktop, mobile, API, or self-hosted environments.

Roboflow suits users seeking a freemium computer-vision platform with a free tier of 10 credits per month, enough for about 30 model trainings or 80,000 inferences.

CVAT is preferable for open-source, self-hosted annotation, with a personal and small-team Community plan and a free online tier limited to one member, one project, three tasks, and 1 GB.

Amazon SageMaker Autopilot is the alternative for pay-as-you-go machine-learning automation without a free plan, but with a free trial.

BRAT is a free, MIT-licensed, self-hosted option.

Intellisane Annotation Platform fits small teams that can work within its free allowance of three members, two projects, 1,000 annotations, and 200 assets.

Label Studio is an alternative freemium labeling platform with API, web, desktop, and self-hosted access.

Supervisely is worth considering for a community tier offering 5 GB, 10,000 files, two members, and unlimited projects and annotations, with a paid Pro tier from €199 a month.

Browse more options in AI Data Labeling Tools.

Verdict

Alibaba Cloud PAI iTAG is best for existing Alibaba Cloud customers that need multimodal templates, custom ontologies, and a reviewer-aware labeling workflow at no platform charge for manual annotation. Its OSS constraints and maintenance-mode, allowlist-only status are the main reasons to look elsewhere, especially if you are a new user building a platform for future growth.