I know others have encountered this bug, but I’m running into it consistently (multiple times a day) in the playground.
And there’s no good workaround – once the error is encountered, the chat isn’t saved in history, so I have to open a new playground window and manually copy-paste each element of the dialogue to re-create it. There’s also nothing I see causing it; I’m simply using GPT-5.2 for coding help.
Hi there,
This error indicates the Playground conversation state got into an invalid sequence for a reasoning model (a reasoning item ended up unpaired with the next required item), so follow-up turns fail and the session may not save. This is a known class of issue with reasoning item chaining in the Responses-backed UI.
As a workaround, you can start a new Playground thread (current one is effectively corrupted).
If you can reproduce, what helps is: browser/OS, whether you were streaming, and whether you used tools (e.g., code execution), plus the rough steps right before it breaks.
I could start a new thread, but the benefit of using GPT in the playground is that I build up a great amount of context over time. Copying all those messages to a new thread is very annoying.
I can’t find patterns in when it breaks, but I use Mac (13.7.6), Chrome browser, with streaming. I’m in the playground all day and it happens 2-3x a day.
I figured it out! It happens when I click “edit” on one of the assistant messages. And it can be resolved by deleting the “reasoning” assistant message right before the one you click “edit” on.
I remember now that it used to give a warning “if you edit this, you’ll lose the reasoning trace”, and I’d say “ok” and move on. But now the reasoning traces are maintained as messages even when an assistant message is edited.
The reason I do this is because I want to copy useful parts of assistant messages and paste them as markdown in local docs. But I can only get the markdown text by hitting “edit” before copying.