json_schema not supported in gpt-chat-latest-2026-05-05 via Responses API

Hello @Sina Salam ,

thanks for your quick response!

I understand everything you say and I am using an API version (2025-03-01-preview), which supports structured output as well as a model (gpt-chat-latest), which is supposed to support structured output (including json_schema, not just json_mode) according to this: https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure?tabs=global-standard&pivots=azure-openai#gpt-chat-latest

Additionally, the model we are currently using (gpt-5.2-chat) is supposed to be retired soon according to this: https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule (btw, I'm not sure, when exactly it will be retired; the German version of that page says 2026-06-09 and the English (US) version says 2026-06-22; also, the German version mentions gpt-5.3-chat as a replacement, which is due to be retired at the same date, while the English version mentions gpt-chat-latest as a replacement).

gpt-5.2-chat supports json_schema nicely and we are using this at the moment without problems. As gpt-chat-latest is listed as a replacement (I would trust the English version more than the German one), I would expect that gpt-chat-latest can cover all features that gpt-5.2-chat offers. But that doesn't seem to be the case at the moment.

So I hope there will be a way to use json_schema in gpt-chat-latest before gpt-5.2-chat is retired.

As a side node: We are on a paid support plan and I got redirected here for a bug report by the Azure Portal. Is this correct? So is this an official support channel? I had thought this was more like a community forum, but if I get help here, that is of course fine for me. Jus let me know, whether I'm in the correct place.

Thanks and best regards,
Steffen

Hello @Steffen Schreiber ,

Thank you for the detailed report and for sharing the exact error message and documentation links. Based on what you described, your concern is valid, and this does not appear to be a simple JSON mode vs. Structured Outputs misunderstanding. The error you are seeing — 400 Invalid parameter: 'text.format' of type 'json_schema' is not supported with model version gpt-chat-latest-2026-05-05 - conflicts with current Microsoft documentation that lists gpt-chat-latest (2026-05-05) as supporting both the Responses API and Structured outputs. - https://learn.microsoft.com/en-us/answers/questions/5899259/json-schema-not-supported-in-gpt-chat-latest-2026, https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure

For clarification, the Foundry Models sold by Azure page shows gpt-chat-latest with support for Chat Completions API, Responses API, and Structured outputs, while the Structured Outputs how-to page does not currently include gpt-chat-latest in its supported-model list. Your runtime result is therefore consistent with a documentation or service inconsistency, not with an obvious issue in your understanding of the feature. - https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure, https://learn.microsoft.com/en-us/azure/foundry/openai/how-to/structured-outputs

Also, your API version choice also appears reasonable. Microsoft documents that Structured Outputs are supported starting with 2024-08-01-preview and remain supported in later preview versions and v1, so using 2025-03-01-preview should not by itself explain this failure. https://learn.microsoft.com/en-us/azure/foundry/openai/how-to/structured-outputs, https://learn.microsoft.com/en-us/azure/foundry/openai/reference

Your migration concern is also well founded. The current Azure OpenAI retirement schedule lists gpt-5.2-chat with a retirement date of 2026-06-22 and names gpt-chat-latest as the replacement. If gpt-chat-latest does not currently accept json_schema in the Responses API, that creates a real migration gap for customers relying on Structured Outputs with gpt-5.2-chat. - https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule, https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure

At this time, the most accurate interpretation is that this is likely one of the following:

• a service-side capability gap,
• a partial rollout issue,
• or a documentation inconsistency between the model catalog and the Structured Outputs support page.

In other words, this does not currently look like a straightforward usage error on your side. -https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure

My recommended next step:

This issue should be validated internally with the Azure OpenAI / Foundry team support so that Microsoft can confirm whether gpt-chat-latest is expected to support json_schema in the Responses API today, or whether the current documentation needs correction. Azure’s support documentation also confirms that more complex issues raised through Microsoft Q&A / Priority Community Support can be escalated when deeper investigation is required. https://learn.microsoft.com/en-us/azure/azure-portal/supportability/priority-community-support, https://learn.microsoft.com/en-us/answers/support/welcome I will escalate right here.

If possible, please share the following details to help support when you're contacted:

• a sanitized sample request showing how text.format / response_format is being sent,
• the deployment name, region, and API version,
• and the request ID / correlation ID from the failed call, if available.

Therefore, just for now. Until this is clarified, the safest workaround is to continue using a model that is explicitly listed as supporting Structured Outputs and is already working in your environment, such as gpt-5.2-chat while it remains available, or another currently listed Structured Outputs model. That is only a temporary workaround, not a resolution to the inconsistency you raised. - https://learn.microsoft.com/en-us/azure/foundry/openai/how-to/structured-outputs, https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule

Regarding your questions:

Yes, Microsoft Q&A is an official Microsoft platform for technical questions. For some Azure support plans, Priority Community Support (PCS) uses Microsoft Q&A as the initial support path, with escalation to a support ticket when needed. Microsoft also distinguishes between Microsoft moderators/staff and Volunteer Moderators, so replies on Q&A are official platform responses, but not every response is necessarily an engineering-level product confirmation on its own. - https://learn.microsoft.com/en-us/answers/support/welcome, https://learn.microsoft.com/en-us/azure/azure-portal/supportability/priority-community-support, https://learn.microsoft.com/en-us/answers/support/moderators

Finally, your expectation of feature parity is reasonable, and the current evidence points to an Azure inconsistency that needs clarification from Microsoft, rather than a misunderstanding in how you are using Structured Outputs.

Cheers

Hello @Sina Salam ,

thanks for escalating the issue. I will happily provide all requested data when support contacts me.

Best regards,
Steffen

Hello @Sina Salam ,

is there anything new regarding this issue? Support has not contacted me so far and if we cannot use gpt-chat-latest when gpt-5.2-chat is shut down on June 22nd we need to test and evaluate a different model well before that date.

I would be happy getting a quick progress report.

Best regards,
Steffen

Looks like I was too quick with my latest comment. json_schema seems to work with gpt-chat-latest now. Thanks for having this fixed!