Applies to: Customers using Azure OpenAI via Custom Integration, or via System Integration with custom prompts.Microsoft is retiring older versions of gpt-4o (v2024-05-13 and v2024-08-06) and gpt-4o-mini on Azure Standard deployments (pay-as-you-go) on March 31, 2026. Use the table below to determine if you need to act.Do I Need to Act?
Your Situation
Model Now Running
Prompt Change Needed?
Next Action
System Integration
N/A
No
None
Provisioned / Global Standard / Data Zone Standard
Unchanged
No
Revisit before Oct 1, 2026
Custom Integration — auto-upgraded
gpt-5.1 / gpt-5-mini
Yes — replace max_tokens
Do it now
Custom Integration — pinned to 2024-11-20
gpt-4o-2024-11-20
No
Revisit before Oct 1, 2026
Custom Integration — no action taken
gpt-5.1 / gpt-5-mini
Yes — replace max_tokens
Do it now
No Action Required
Scenario
Reason
You use System Integration with System Prompts for Azure OpenAI
Our platform manages your endpoint automatically. No changes needed.
If your deployment type is Global Standard, Provisioned, or Data Zone Standard
Action Required—Custom Integration or Custom Prompts
If you configured Azure OpenAI using Custom Integration, or used custom prompts via System Integration, act based on your current situation. First, check your model version in the Microsoft AI Foundry Portal.
Applies if your update policy is set to Auto-update in Microsoft AI Foundry, or if your older model version retired.Your model is now running gpt-5.1 instead of gpt-4o and gpt-5-mini instead of gpt-4o-mini.Update all custom prompts that use the max_tokens parameter to replace max_tokens with max_completion_tokens. The gpt-5 series of models don’t support max_tokens. If you haven’t updated, you see this error: "Unsupported parameter: 'max_tokens' is not supported with this model. Use 'max_completion_tokens' instead."
Scenario B—You manually pinned your deployment to gpt-4o-2024-11-20
Applies if you updated your model version to 2024-11-20 in Microsoft AI Foundry before the retirement deadline.No action required immediately. This version supports both max_tokens and max_completion_tokens — no prompt changes needed. Update your deployment before October 1, 2026, when this version is also retired.
Scenario C—You took no action and auto-update was off
Microsoft updates your deployment after their model retirement. Your model runs on gpt-5.1 or gpt-5-mini after the update. Follow the steps in Scenario A.See Microsoft AI Foundry Model Retirements.
The Platform displays a notice for NLP-based apps to inform users of the upcoming retirement of select network types and embedding models. Supported network types include Standard, Multi-lingual, Zero-shot, and Few-shot. Supported embeddings include Pre-trained MPNet and BGE M3.
The latest release of Search AI introduces important updates to improve the performance, stability, and flexibility of the platform. As part of this upgrade, some older components and models are being deprecated to make way for new and improved features.
BGE V2 - For text embeddings (higher accuracy than all legacy models)
VDR - For Image embeddings
Impact on Existing ApplicationsAfter November 14, your existing applications will automatically switch to BGE V2 and VDR.What will Change
Search results may rank differently due to improved semantic understanding.
Higher accuracy in finding relevant content, especially for complex queries.
Better handling of domain-specific terminology and context.
Why This Change Benefits You
25-40% improvement in search relevance across diverse domains.
Better semantic understanding reduces the need for query optimization.
The unified model approach ensures consistent performance across text types.
How to Check Your Current Configuration
Go to your application > Search AI > Vector Configuration
Check the embedding model listed.
For applications already using BGE, to determine the version being used, refer to the application creation date. Apps created before the dates listed below use BGE-M3 V1, and apps created after these dates automatically use BGE V2, if BGE is selected.
US region: after July 15, 2025
Japan and Germany: after July 21, 2025
EU: after July 30, 2025
Australia and UAE: after July 31, 2025
India region: after August 7, 2025
Please note that if you have created an app after the date mentioned above, you don’t need to create a new app to try out BGE V2.How to Test Before Migration
Create a new application with the same source content as your existing app.
Your app will be automatically enabled with the BGE V2 model for embeddings.
Run identical searches and compare the results side by side. Specifically test the edge cases that are important to your use case.
How to UpgradeGo to the Vector Configuration page of your Search AI app.
If your app was created after the above-mentioned dates and is still using a deprecated model, select BGE M3 or VDR Vector Model from the dropdown, as appropriate.
For older apps, created before the above-mentioned dates, the transition will occur automatically according to the deprecation and migration timelines.
Impact on Existing ApplicationsIf your application is still using the old crawler, we will automatically migrate it to the new crawler and initiate a re-crawl of the configured webpages. All the functionality stays the same between the two crawlers, so we will honor your existing configurations and re-crawl the content using the new crawler.What will Change
Crawl times reduced by 80%.
Better extraction from JavaScript-heavy sites.
Cleaner content with automatic filtering of navigation, ads, and boilerplate text.
More successful crawls of complex websites.
Why This Change Benefits You
Faster time-to-value with dramatically reduced crawl times
Higher success rates on complex, modern websites
Better content quality with intelligent filtering of irrelevant text
Reduced resource usage means lower costs and better platform stability
Note: The new crawler was released on May 31, 2025. New applications automatically use it, while existing applications can upgrade early via banner notification or will be automatically migrated by November 14.How to UpgradeIf your app is using the old crawler, you’ll see an upgrade banner on the Websites page under Content.Click on the banner to start the upgrade process.Click on Switch to New Crawler. The crawler is automatically updated to the new crawler.
MS Macro Cross Encoder (sentence-transformers/all-MiniLM-L6-v2)
Mixbread Large (mixedbread-ai/mxbai-rerank-large-v1)
Supported Model (Default):
BGE (BAAI/bge-reranker-v2-m3) - Provides the highest accuracy among all re-ranker options
Impact on Existing ApplicationsIf you’re currently using re-ranker models, your applications will automatically switch to BGE:What will change:
Applications using MS Macro Cross Encoder or Mixbread Large will automatically use BGE-M3.
Improved re-ranking accuracy, especially for complex queries requiring nuanced understanding.
Consistent performance across different content types and languages.
If you’re not using re-ranker models, there is no impact.How to UpdateIf you are using a re-ranker, to update the re-ranker model, follow the instructions.
Go to your application > Search AI > Advanced Configuration
Set up BGE-M3 re-ranker in Advanced Configuration → Re-Rank Chunks
The existing applications will be automatically migrated in batches by November 14, 2025. Upon completion of the migration, you will receive a pop-up notification confirming the upgrade.
Date
Action
Now
Test new features by creating new applications.
November 14, 2025
Automatic migration of all existing applications.
After November 14, 2025
Legacy models and old crawlers will no longer be available.
Can I opt out of this migration?No, this is a platform-wide upgrade that will apply to all applications to ensure consistent performance and enable new features.Will my search results change?You may notice improvements in search accuracy due to the superior BGE V2 model. Some specific result rankings may change, but overall relevance will improve.Can I still use the old crawler?No, after November 14, 2025, only the new crawler will be available. However, the new crawler provides significant performance and reliability improvements.What if my application breaks after migration?While we expect the migration to be seamless, our support team will be available to assist with any issues that may arise. We recommend testing with a new application beforehand to identify any potential concerns.Will I be charged differently for the new models?Pricing remains the same. You’re getting improved performance and accuracy at no additional cost.
