Azure AI Vision, still called the Computer Vision API in its REST endpoints, is Microsoft’s cloud service for analysing images and video: tagging, captions, object detection, OCR and people detection. Microsoft now groups it under Foundry Tools as “Azure Vision in Foundry Tools” (Microsoft Learn).
Updated October 2026
What changed in 2026
- New name. Microsoft now calls the service Azure Vision in Foundry Tools, part of Microsoft Foundry (Microsoft Learn).
- Retirement announced. The Image Analysis API (v3.2 and v4.0, cloud and containers) retires on 25 September 2028, and Microsoft asked customers to have a transition plan by 25 September 2026 (Microsoft Learn).
- Custom Vision retires on the same date. Microsoft points to AutoML in Azure Machine Learning, Foundry models or Content Understanding (preview) instead (Microsoft Learn).
- Old Computer Vision APIs stopped. Versions 1.0 to 3.1 retired on 13 September 2026 (Microsoft Learn).
- Replacements are generative. OCR moves to the Document Intelligence Read model; general image analysis moves to GPT models in Foundry or Azure Content Understanding (Microsoft Learn).
Names have changed
| Older name | Current name |
|---|---|
| Cognitive Services | Azure AI services, now Foundry Tools |
| Computer Vision | Azure AI Vision / Azure Vision in Foundry Tools |
| Form Recognizer | Azure AI Document Intelligence |
| Custom Vision | Azure AI Custom Vision |
What it does
- Image Analysis 4.0 (GA since November 2023): tags, captions (which replaced “Describe”), dense captions, object and people detection, smart crops.
- Read (OCR): printed and handwritten text from images and documents.
- Custom Vision: train your own classifiers and object detectors, now that customisation inside Image Analysis has been removed. Custom Vision itself is scheduled to retire on 25 September 2028, so new projects should look at the replacements Microsoft recommends.
- Face: detection and verification under Limited Access. Emotion, gender and age inference was withdrawn in 2022 (Microsoft).
Dates to plan for
| What | Date |
|---|---|
| Customisation, background removal, product recognition removed from Vision Studio | 10 January 2025 |
| Image Analysis 4.0 preview APIs retired | 31 March 2025 |
| Computer Vision API v1.0, 2.0, 3.0 and 3.1 retired; calls stop working | 13 September 2026 |
| Transition plan requested by Microsoft | 25 September 2026 |
| Image Analysis (v3.2 and v4.0) and Custom Vision retire | 25 September 2028 |
If your apps still call v3.1 or earlier, those calls have stopped since 13 September 2026. Moving to v3.2 or Image Analysis 4.0 restores service for now, but both retire in September 2028, so plan the longer move to Foundry models, Content Understanding or Document Intelligence at the same time.
Typical uses
| Industry | Use |
|---|---|
| Retail | Product tagging, shelf images, visual search |
| Manufacturing | Basic defect classification with Custom Vision |
| Documents | Reading invoices, forms and IDs (with Document Intelligence) |
| Accessibility | Image captions and alt text |
Azure AI Vision vs AWS Rekognition vs Google Cloud Vision
All three offer labelling, OCR and object detection as pay-per-call cloud APIs. Choose mainly by where your data and other services already run. For video at many sites, low latency, or footage that must stay on-premises, a cloud-only API is not enough; models need to run at the edge.
Where NexaStack fits
Azure AI Vision tells you what is in an image. NexaStack runs vision models, including Azure ones, at the edge or in your cloud. It connects the results to AI agents that act in your systems, under rules and approvals you set. See computer vision applications and NexaStack Vision AI.
Where it's heading: 2026 → 2030
| When | What it looks like |
|---|---|
| Around 2020 | Computer Vision API 3.x and Custom Vision under the Cognitive Services brand: fixed-feature calls for tags, OCR and objects. |
| 2026 (now) | Azure Vision in Foundry Tools still works, but Microsoft has announced its retirement and steers new work to Foundry models, Content Understanding and Document Intelligence. |
| By 2028 (expected) | Image Analysis and Custom Vision calls fail after 25 September 2028; apps must have moved (Microsoft Learn). |
| By 2030 (expected) | Vision on Azure runs mainly through multimodal models in Foundry in the cloud, with smaller models on edge devices for latency and data-residency needs. This is our expectation, not a Microsoft roadmap. |
Frequently asked questions
Is Azure Computer Vision being discontinued?
In stages. Versions 1.0 to 3.1 stopped working on 13 September 2026, and Microsoft has announced that Image Analysis (v3.2 and v4.0) and Custom Vision retire on 25 September 2028. New work should start on Foundry models, Content Understanding or Document Intelligence.
Can Azure detect emotions from faces?
No longer. Microsoft withdrew emotion, gender and age inference in 2022.
Can Azure AI Vision run offline?
Some capabilities, such as Read (OCR), are offered as containers that run on your own hardware, subject to Microsoft's approval and billing rules. Check the current container list on Microsoft Learn, as the Image Analysis containers follow the same 2028 retirement.
What are the latest Azure AI Vision changes in 2026?
The service is now Azure Vision in Foundry Tools, its Image Analysis and Custom Vision APIs are set to retire on 25 September 2028, and Microsoft recommends Foundry models, Content Understanding and Document Intelligence as replacements.
Plan your move off Azure Vision APIs
Retirement dates are set. We can review which of your vision workloads should move to newer models and which should run at the edge. Book a 30-minute vision migration review →
