Starting this morning, the Chat Completion API has begun returning malformed or inconsistent JSON. I am using gpt-4o-2024-11-20 with JSON-style responses. My prompts have not changed, but the model is now intermittently returning JSON objects with: Missing dictionary keys Keys placed at the w…

I am getting degraded outputs right now too! I think there might be an issue

Definitely not resolved. I thought that having an Azure deployment with a date is some sort of a promise but appears like its broken. We switched about 60% of our flows to Gemini, and still seeing crazy logs and traces from the remaining 40% we still haven’t tested and moved. Maybe this has to do…

Im using direct conn to openai not azure. but this is really disapointing. the model has been fine for over a year.

and using the same ft model (not some changed snapshot of a foundation model). so not sure how it changed…. must be the software layer - haven’t even changed my version!

Yes, I can bring this up with the team. We’ll have to wait and see what happens next. For now and the future the best recommendation is to switch to Structured Outputs. JSON mode has had recurring issues, and that’s exactly why Structured Outputs were introduced.

Structured outputs also suffer from the same issue. BTW the internal mechanism for structured ouputs and json mode is the same (zero-out all the log-probs of the tokens that violate the schema)

To escalate a this issue with json-mode or structured outputs, it really helps the team if you can share a few response IDs. That allows them to inspect and reproduce the problem directly. Also, keep in mind that structured outputs work differently from JSON mode. JSON mode relies on prompting plus…

I’m using .NET and the Responses API - any pointer on docs to set this up? Happy to switch to structured out as the system is getting built at the moment. Switch code from assistant API where everything was working great - this situation is not idea :frowning: A bit of a moan - but the .NET NuGet pa…

Thanks for assisting Json failure: response id: chatcmpl-Cla2GyONCawwxYxXdzFyHId6Tauim For the Structured output failure: request_id: fa3168bc-496d-4b21-9652-d7fe09370d82 I couldn’t retrieve the response id conveniently, since it is not returned from the python openai package, because it fails.…

I will send some IDs as well. I’m getting issues with structured outputs not json mode.

Chat Completion responses suddenly returning malformed or inconsistent JSON

API Bugs

roy.amit December 11, 2025, 4:47pm 18

You can just make some 200 calls asking for a json, it will fail 1-5 times.

Topic		Replies	Views
Degraded Performance on gpt-4.1-mini ft Bugs chat-completion , structured-output	28	771	January 25, 2026
Response has valid json but it's nested in broken json Bugs	17	4398	December 26, 2025
Function calling looping uncontrollably and calling unnecessarily Bugs function-calling , gpt-4o , gpt-4o-mini	27	4323	September 19, 2024
Structured Outputs not reliable with GPT-4o-mini and GPT-4o API structured-output	38	9696	January 23, 2025
Responses API returning \t\t\n\n \t\t\n \t\t\n \t\t\n Bugs responses	17	2173	December 1, 2025

Chat Completion responses suddenly returning malformed or inconsistent JSON

Related topics