Introducing the Responses API

Announcements
23

Despite “continued support” stated, this sounds like Chat Completions deprecation to me, when new models can no longer be used…

image
image566×103 5.9 KB

1 Like
24

Is Response API available for all users yet? I can’t access it with my API key, though it has access to gpt-4o and gpt-4o-mini. It returns 403 forbidden “Project … does not have access to model gpt-4o-mini…”, even though with the same API key I can hit chat completion using the same models.

25

API keys can have allowed endpoints set on them.

You will note that there is no ‘/responses’ showing here in “Restricted” where you must enable individual endpoints:

The image shows a settings interface for editing a secret key, with options to set permissions for different resources such as models and model capabilities. (Captioned by AI)
The image shows a settings interface for editing a secret key, with options to set permissions for different resources such as models and model capabilities. (Captioned by AI)426×488 15.8 KB

An “All” API key was immediately applicable to the Responses endpoint for me.

You may need to simply generate a new API key to be placed in your application for the endpoints behind “all” to be properly populated.

26

Thanks for your reply! Yes I set the permission to “All”, but I can’t seem to hit the endpoint successfully. I’ll try your suggestion, if it’s not succeed I’ll start another thread. Thank you! :grinning_face_with_smiling_eyes:

1 Like
27

Any ETA for code interpreter for Responses API?

3 Likes
28

Is there a Completions to Responses migration guide? I couldn’t find one … that would be useful.

However, in any case I’m not 100% a fan of switching unless Responses becomes an industry standard, because there are circumstances where you want to be able to support both OpenAI and third parties services, e.g. in generic open source repos.

Today I might want to use GPT 4.1, but tomorrow I might need to switch to Meta’s Llama 4 perhaps, something I can do with far less friction with Completions.

1 Like
29

https://platform.openai.com/docs/guides/responses-vs-chat-completions?api-mode=responses#compare-the-code

3 Likes
30

Great find, thanks, very useful!

31

I understand about not wanting to switch and it’s hard to justify any reason. You get an internal iterator and response id storage for:

  • server-side chat state: that can run a new model up to its 1M on chat with no management of the length;
  • file search that has injected messages saying “user uploaded files”, destroying most use-case;
  • web search that takes control and only produces a dump.
  • reasoning “thinking” - then blocked without personal ID verification

…to then take away logprobs, logit_bias, etc.

4 Likes
32

I think the one benefit you get from the Responses API (if just doing single transaction “completions” and local tools) is reasoning summaries?

1 Like
33

This is the discussion I have been looking for, thank you all for your insights!:flexed_biceps:

1 Like
34

They added a selector in the right panel for an equivalency of code for both versions in each api docs:

The image provides an overview of using the OpenAI API for text generation, featuring a code example with the Responses API and a dropdown menu showing different API options. (Captioned by AI)
The image provides an overview of using the OpenAI API for text generation, featuring a code example with the Responses API and a dropdown menu showing different API options. (Captioned by AI)1492×678 86.1 KB

The image shows a webpage titled "Images and vision" that describes vision capabilities for understanding images, including instructions for using URLs or Base64 encoded images. (Captioned by AI)
The image shows a webpage titled "Images and vision" that describes vision capabilities for understanding images, including instructions for using URLs or Base64 encoded images. (Captioned by AI)1475×807 90.3 KB

2 Likes
36

Myself and my team are working on a middleware server.
It lets you use Responses API (+features! We’re working on Web search, file search, etc.) but with any chat-completions compliant AI Provider.

It’s called “openai-responses-server” on GitHub (Can’t include links..)
And we’re adding it as pypi package or docker image.

37

Any update on the timeline for Code interpreter? This is a key feature keeping us from switching to the Responses API

1 Like
38

Hi, I am still waiting for “previous_response_id” (as a way to preserve conversation) to handle truncation of the chain based on message date/time , max number, or max tokens. With the release of models handling 1M tokens, this is even more important to control costs.

1 Like