[FEATURE REQ] Make escape hatches available in Azure OpenAI

[janaka]

Library name

Azure.AI.OpenAI

Please describe the feature.

Please review your design strategy to increasingly seal off all the escape hatches available in the OpenAI SDK. We are adults and don't need protecting in this way - if that's the goal. By all means provide conveniences for the happy paths but also make it possible for consumers to drop down the layers, inherit, override, and extend making the full range of options available for working around problems.

By taking this approach you have created an unnecessary hard dependency on the Azure OpenAI SDK team. And hard blocking many issues/features ongoing that could other wise could be worked around.

Examples of real impact:

• protocol methods are all sealed off in the Azure OpenAI SDK therefore it cannot be used to workaround issues like [BUG] gpt-image-1 model does not work with latest API version 2.2.0-beta4 #49965 where a parameter in the request body needs to be adjusted.

• ServiceVersion enum used to be able to override using reflection because the property was a string, but that has changed so this method of workaround doesn't work. ideal there would be a first class way of passing the version number as a string. Again the lack of escape hatch is preventing a simple solution to the problem in issue [BUG] gpt-image-1 model does not work with latest API version 2.2.0-beta4 #49965 rather I've had to drop to the HTTP layer and write a policy that intercepts the request and rewrites the API version param.

[github-actions]

Thanks for the feedback! We are routing this to the appropriate team for follow-up. cc @jpalvarezl @ralph-msft @trrwilson.

[github-actions]

Hi @janaka, I'm an AI Support assistant here to help with your issue. While the Azure SDK team reviews your request, I wanted to provide some possible tips and documentation that might help you in the meantime.

• The current design of the Azure OpenAI SDK means that many internal operations are intentionally sealed to maintain consistency and simplify the “happy path.”
• We’re aware that this design can limit certain workarounds—for example, adjusting protocol parameters or version strings (as noted in issue [BUG] gpt-image-1 model does not work with latest API version 2.2.0-beta4 #49965).
• Some of our recent releases (see Azure.AI.Inference CHANGELOG) illustrate our commitment to evolving the SDK while preserving a well-defined API surface.
• As noted in other parts of our changelogs (for instance, Azure.ResourceManager.Advisor CHANGELOG), we continue to review user feedback—even on scenarios requiring more advanced customization. The evolution of these packages shows that we’re actively considering the balance between simplicity and extensibility.

The Azure SDK team will respond to your issue shortly. I hope these suggestions are helpful in the meantime. If this comment helped you, please give it a 👍. If the suggestion was not helpful or incorrect, please give it a 👎. Your feedback helps us improve!