Edit

What's new in Azure OpenAI Service

  • Article

September 2023

GPT-4

GPT-4 and GPT-4-32k are now available to all Azure OpenAI Service customers. Customers no longer need to apply for the waitlist to use GPT-4 and GPT-4-32k (the Limited Access registration requirements continue to apply for all Azure OpenAI models). Availability may vary by region. Check the models page, for the latest information on model availability in each region.

GPT-3.5 Turbo Instruct

Azure OpenAI Service now supports the GPT-3.5 Turbo Instruct model. This model has performance comparable to text-davinci-003 and is available to use with the Completions API. Check the models page, for the latest information on model availability in each region.

Whisper public preview

Azure OpenAI Service now supports speech to text APIs powered by OpenAI's Whisper model. Get AI-generated text based on the speech audio you provide. To learn more, check out the quickstart.

Note

Azure AI Speech also supports OpenAI's Whisper model via the batch transcription API. To learn more, check out the Create a batch transcription guide. Check out What is the Whisper model? to learn more about when to use Azure AI Speech vs. Azure OpenAI Service.

New Regions

  • Azure OpenAI is now also available in the Sweden Central, and Switzerland North regions. Check the models page, for the latest information on model availability in each region.

Regional quota limits increases

  • Increases to the max default quota limits for certain models and regions. Migrating workloads to these models and regions will allow you to take advantage of higher Tokens per minute (TPM).

August 2023

Azure OpenAI on your own data (preview) updates

July 2023

Support for function calling

Embedding input array increase

New Regions

  • Azure OpenAI is now also available in the Canada East, East US 2, Japan East, and North Central US regions. Check the models page, for the latest information on model availability in each region.

June 2023

Use Azure OpenAI on your own data (preview)

  • Azure OpenAI on your data is now available in preview, enabling you to chat with OpenAI models such as GPT-35-Turbo and GPT-4 and receive responses based on your data.

New versions of gpt-35-turbo and gpt-4 models

  • gpt-35-turbo (version 0613)
  • gpt-35-turbo-16k (version 0613)
  • gpt-4 (version 0613)
  • gpt-4-32k (version 0613)

UK South

  • Azure OpenAI is now available in the UK South region. Check the models page, for the latest information on model availability in each region.

Content filtering & annotations (Preview)

Quota

May 2023

Java & JavaScript SDK support

  • NEW Azure OpenAI preview SDKs offering support for JavaScript and Java.

Azure OpenAI Chat Completion General Availability (GA)

  • General availability support for:
    • Chat Completion API version 2023-05-15.
    • GPT-35-Turbo models.
    • GPT-4 model series.

If you are currently using the 2023-03-15-preview API, we recommend migrating to the GA 2023-05-15 API. If you are currently using API version 2022-12-01 this API remains GA, but does not include the latest Chat Completion capabilities.

Important

Using the current versions of the GPT-35-Turbo models with the completion endpoint remains in preview.

France Central

  • Azure OpenAI is now available in the France Central region. Check the models page, for the latest information on model availability in each region.

April 2023

  • DALL-E 2 public preview. Azure OpenAI Service now supports image generation APIs powered by OpenAI's DALL-E 2 model. Get AI-generated images based on the descriptive text you provide. To learn more, check out the quickstart. To request access, existing Azure OpenAI customers can apply by filling out this form.

  • Inactive deployments of customized models will now be deleted after 15 days; models will remain available for redeployment. If a customized (fine-tuned) model is deployed for more than fifteen (15) days during which no completions or chat completions calls are made to it, the deployment will automatically be deleted (and no further hosting charges will be incurred for that deployment). The underlying customized model will remain available and can be redeployed at any time. To learn more check out the how-to-article.

March 2023

  • GPT-4 series models are now available in preview on Azure OpenAI. To request access, existing Azure OpenAI customers can apply by filling out this form. These models are currently available in the East US and South Central US regions.

  • New Chat Completion API for GPT-35-Turbo and GPT-4 models released in preview on 3/21. To learn more checkout the updated quickstarts and how-to article.

  • GPT-35-Turbo preview. To learn more checkout the how-to article.

  • Increased training limits for fine-tuning: The max training job size (tokens in training file) x (# of epochs) is 2 Billion tokens for all models. We have also increased the max training job from 120 to 720 hours.

  • Adding additional use cases to your existing access.  Previously, the process for adding new use cases required customers to reapply to the service. Now, we're releasing a new process that allows you to quickly add new use cases to your use of the service. This process follows the established Limited Access process within Azure AI services. Existing customers can attest to any and all new use cases here. Please note that this is required anytime you would like to use the service for a new use case you did not originally apply for.

February 2023

New Features

Updates

  • Content filtering is on by default.

New articles on:

New training course:

January 2023

New Features

  • Service GA. Azure OpenAI Service is now generally available.​

  • New models: Addition of the latest text model, text-davinci-003 (East US, West Europe), text-ada-embeddings-002 (East US, South Central US, West Europe)

December 2022

New features

  • The latest models from OpenAI. Azure OpenAI provides access to all the latest models including the GPT-3.5 series​.

  • New API version (2022-12-01). This update includes several requested enhancements including token usage information in the API response, improved error messages for files, alignment with OpenAI on fine-tuning creation data structure, and support for the suffix parameter to allow custom naming of fine-tuned jobs. ​

  • Higher request per second limits. 50 for non-Davinci models. 20 for Davinci models.​

  • Faster fine-tune deployments. Deploy an Ada and Curie fine-tuned models in under 10 minutes.​

  • Higher training limits: 40M training tokens for Ada, Babbage, and Curie. 10M for Davinci.​

  • Process for requesting modifications to the abuse & miss-use data logging & human review. Today, the service logs request/response data for the purposes of abuse and misuse detection to ensure that these powerful models aren't abused. However, many customers have strict data privacy and security requirements that require greater control over their data. To support these use cases, we're releasing a new process for customers to modify the content filtering policies or turn off the abuse logging for low-risk use cases. This process follows the established Limited Access process within Azure AI services and existing OpenAI customers can apply here.​

  • Customer managed key (CMK) encryption. CMK provides customers greater control over managing their data in Azure OpenAI by providing their own encryption keys used for storing training data and customized models. Customer-managed keys (CMK), also known as bring your own key (BYOK), offer greater flexibility to create, rotate, disable, and revoke access controls. You can also audit the encryption keys used to protect your data. Learn more from our encryption at rest documentation.

  • Lockbox support​

  • SOC-2 compliance​

  • Logging and diagnostics through Azure Resource Health, Cost Analysis, and Metrics & Diagnostic settings​.

  • Studio improvements. Numerous usability improvements to the Studio workflow including Azure AD role support to control who in the team has access to create fine-tuned models and deploy.

Changes (breaking)

Fine-tuning create API request has been updated to match OpenAI’s schema.

Preview API versions:

{​
    "training_file": "file-XGinujblHPwGLSztz8cPS8XY",​
    "hyperparams": { ​
        "batch_size": 4,​
        "learning_rate_multiplier": 0.1,​
        "n_epochs": 4,​
        "prompt_loss_weight": 0.1,​
    }​
}

API version 2022-12-01:

{​
    "training_file": "file-XGinujblHPwGLSztz8cPS8XY",​
    "batch_size": 4,​
    "learning_rate_multiplier": 0.1,​
    "n_epochs": 4,​
    "prompt_loss_weight": 0.1,​
}

Content filtering is temporarily off by default. Azure content moderation works differently than OpenAI. Azure OpenAI runs content filters during the generation call to detect harmful or abusive content and filters them from the response. Learn More​

​These models will be re-enabled in Q1 2023 and be on by default. ​

​Customer actions​

Next steps

Learn more about the underlying models that power Azure OpenAI.