As of this morning with no changes no our end, every model in the playground returns “Encrypted content is not supported with this model.” regardless of the prompt or user input.
I am facing the same problem.
Same issue here. I just typed in “Test” and get the same error. Cleared Cookies and Cache and it did not help.
I also have the same issue.
Yep, it’s screwed up. But not if you do use a reasoning model like o4-mini.
The playground is sending an empty reasoning object, which like sending tools, will get a block on a non-tools model, and here is likely going to have implication of a request for a default reasoning level.
This could be an API issue and behavior change there also.
Use Chat Completions, receive no nonsense except the intentional breaking of the presets features.
I just checked a tab with a playground open 24 hours + (still have to get back to the results)
- the get code snippet shows the reasoning object still
- no errors
So it seems something more.
I had to look at the actual request made, not the “get code”
They changed the playground code. It is sending "include":["reasoning.encrypted_content"]} to all models.
This is hopefully the start of fixing the bad stateless behavior, so reasoning models don’t fail with out using store.
Yes: the request fulfillment on a reasoning model is now adding the encrypted content from AI output to even the “get code” messages.
I report having the exact same issue.
I’m having the same issue.
I’m not seeing this issue reflected there.
Much better would be if the API would just drop any non-supported “include”.
With use of their “prompts” for server side settings, but not everything you need to know, and no way to actually get it, you’d have no idea if a prompt can support sending a include request for code interpreter outputs, for example, or indeed if there is a reasoning model involved that can support encrypted reasoning that you must ask for and can only ask for when store: false.
Anyone have tips for getting OpenAI’s attention? Seems like a pretty major issue here.
How is this still an issue? Does anyone actually work there?
Hard-refresh the platform site. The playground issues seem to have been repaired, deciding if the model is reasoning before including a request for encrypted reasoning or it was reverted, in that GPT-3.5 is sending an empty include array to the API now, along with an empty reasoning object.
Confirming @_j here. This was fixed yesterday. If you’re still seeing issues, do a hard refresh. Sorry for the troubles!