Prompt playground no longer works

[john118]

All at once about an hour ago, my entire team gets the following error when trying to test a prompt in the prompt playground:

Unsupported content type: ‘text/plain’. This API method only accepts ‘application/json’ or ‘application/x-www-form-urlencoded’ requests, but you specified the header ‘Content-Type: text/plain’. Please try again with a supported content type.

[Foxalabs]

Hi,

Which model were you testing? can you show the setting being used at the time?

[john118]

This is occurring across several models and settings - we’re unable to change the model and settings to prevent it from happening. But here is one example:

[Sameer_Ahmed]

happening with me with exact same error and it is not related to any extension or any thing

[_j]

The error message indicates the headers in the API request that the Playground makes are not being created correctly.

I have a browser tab that has been open since yesterday, and it is able to submit the same request to Chat Completions.

Replication: I gave a hard-refresh with the shift key, and then with ctrl key to reload as a new tab. Switched to “Responses”. Of the two “prompts” as presets I’ve copied from “presets”, then dealt with, “The model “gpt-4.5-preview” is not available” because they turned off the model desired. The “default model” is then gpt-4.1 like you show.

Verification: The request fails. (no tools)

The API request is indeed sending screwed-up headers.

POST /v1/responses HTTP/3
Host: api.openai.com
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:141.0) Gecko/20100101 Firefox/141.0
Accept: text/event-stream
Accept-Language: en-US,en;q=0.5
Accept-Encoding: gzip, deflate, br, zstd
Referer: https://platform.openai.com/
Authorization: Bearer sess-
OpenAI-Organization: org-
OpenAI-Project: proj_
OpenAI-Beta: responses=v1
Content-Type: text/plain;charset=UTF-8
Content-Length: 356
Origin: https://platform.openai.com
Connection: keep-alive
Sec-Fetch-Dest: empty
Sec-Fetch-Mode: cors
Sec-Fetch-Site: same-site
Priority: u=0
Pragma: no-cache
Cache-Control: no-cache
TE: trailers

Content-Type: text/plain;charset=UTF-8, sending JSON - Smashing brilliance.

Solution: use Chat Completions, still working. Abandon Responses permanently.

@_j

Degrade and remove useful models and foist a continually-broken endpoint by any means of coercion and vendor lock-in, not going to work.

[john118]

Sadly the chat completions API does not support all the functionality we need.

[Foxalabs]

Thanks for taking the time to report this, I’ve passed it to OpenAI.

[_j]

@john118

chat completions API does not support all the functionality

If it is due to web search that you need Responses:

• make a function for “web search”. Describe it how and when you actually want web search utilized.
• Your function can then return web search results without any language injection demanding that web search take over the output, giving only short snippets with links and citations to go to actual sites like search results. It can actually inform the AI with knowledge and not be a results page.
• You can use any web search API,
• or if you want a web “product” as RAG, you can use the chat completions web search model as a function, which only costs you one call instead of going into loops and always searches.

Of course, you can’t test that on the playground, but you can develop a better application.

[Foxalabs]

I tried without websearch, still has the same error.

[dmitry-p]

Thanks for flagging, we’re about to deploy a fix. Apologies again!

Update: should be deployed

[_j]

Refresh/reload for working “Responses” API endpoint in “Prompts” with “prompt” and then sending your prompt.

…to receive a prompt.