This browser is no longer supported.
Upgrade to Microsoft Edge to take advantage of the latest features, security updates, and technical support.
![](data:image/svg+xml, %3Csvg xmlns='http://www.w3.org/2000/svg' height='64' class='font-weight-bold' style='font: 600 30.11764705882353px "SegoeUI", Arial' width='64'%3E%3Ccircle fill='hsl(70.4, 41%, 16%)' cx='32' cy='32' r='32' /%3E%3Ctext x='50%25' y='55%25' dominant-baseline='middle' text-anchor='middle' fill='%23FFF' %3EK%3C/text%3E%3C/svg%3E)
We've deployed GPT-4.1 on Azure AI Foundry (OpenAI). We're using the Responses API (https://free.blessedness.top/en-us/azure/ai-foundry/openai/reference-preview-latest#create-response) with structured outputs (a JSON schema), which has been working fine for the past couple of months.
We've suddenly started getting this error: Error code: 400 - {'error': {'message': 'There was an issue with your request. Please check your inputs and try again', 'type': 'invalid_request_error', 'param': None, 'code': None}}
After some debugging, we saw that this error is caused by the JSON schema we're using (it worked with a very simple JSON schema). However, we've been using our same JSON schema for the past couple of months. Even after simplifying it and drastically shortening it, we got the same error.
It's also working fine with OpenAI's API.
Has anything changed with how the Azure OpenAI API handles structured outputs in the past 24 hours (7-8 October 2025)?
![](data:image/svg+xml, %3Csvg xmlns='http://www.w3.org/2000/svg' height='64' class='font-weight-bold' style='font: 600 30.11764705882353px "SegoeUI", Arial' width='64'%3E%3Ccircle fill='hsl(316.8, 39%, 40%)' cx='32' cy='32' r='32' /%3E%3Ctext x='50%25' y='55%25' dominant-baseline='middle' text-anchor='middle' fill='%23FFF' %3ESM%3C/text%3E%3C/svg%3E)
Hi Kab,
There have been recent updates to the Azure OpenAI Responses API that may be affecting your schema compatibility:
1**.Stricter Schema Validation Introduced**
Azure OpenAI now enforces stricter validation rules for structured outputs. Even minor deviations from expected schema formats such as missing operationId, placeholder URLs, or incomplete metadata can trigger a 400 Bad Request.
2**.Model and API Version Compatibility**
Certain models (like gpt-4.1, gpt-4o, gpt-5) now require specific API versions (e.g., 2024-10-21, 2025-03-01-preview) for structured outputs to work correctly. If you're using an older version, even a valid schema May failturn1search4.
3.Known Issues with response_format
There are reports of intermittent failures with structured outputs, especially when using response_format with Pydantic classes. These errors may occur after a few minutes of successful operation, suggesting a backend regression or timeout issue.
To Resolve the issue:
Ensure API Version Compatibility
Use the latest supported API version for your model. For example:
client = AzureOpenAI(
azure_endpoint=os.getenv("AZURE_OPENAI_ENDPOINT"),
azure_ad_token_provider=token_provider,
api_version="2025-03-01-preview"
)
Validate Your JSON Schema Strictly
Test with Minimal Schema First
Check for Known Limitations
Structured outputs are not supported with:
https://free.blessedness.top/en-us/azure/ai-foundry/openai/how-to/responses?tabs=python-key
I Hope this helps. Do let me know if you have any further queries.
Thank you!
Thank you @Sridhar M.
As it's working with a smaller JSON output schema, it doesn't look like the issue is #2. FWIW, we're using the API version preview. Hopefully it's #3 and it's temporary (although it's been persisting several hours).
On point #1, I understand the point on strict schema validation enforcement, but that JSON schema is working with OpenAI's native API, using GPT-4.1. When was the recent update that has enforced stricter validation than OpenAI's API, and where can I find documentation on it please?
This issue seems to have been resolved on its own, thanks @Sridhar M.
However this significantly reduces confidence in Azure OpenAI model deployments being ready for production. It would be great for other customers as well to understand why this happened.