The Chat Playground is now the Prompts Playground

https://x.com/OpenAIDevs/status/1900358343333482861

You can also save and share specific configurations of models, system messages, and tools. For example, here’s a starter prompt with gpt-4o-mini using the Responses API with web search: https://platform.openai.com/playground/p/T4U6rwVjng3kKl4UjaPr2CCZ?mode=chat

Long live Prompts, now sending “inputs”! Where you can use Responses to see responses!

The ability to reorganize context and thoughtful use of space for parameters is well done, even if everyone will have their own feedback.


And a farewell, in two days, to the “completions” Playground, returning a visualization of logprob certainties, results, and perplexity, explorations of the science behind language models - and what “prompt engineering” used to actually mean on a completions endpoint “engine”.

(Completions Playground is now shut off: Only API calls)

Prompts Playground needs a reset button, to clear all content, tools and responses, and the association with a preset.

But maintain the model and parameters.

A bookmark to the base URL can do this, a slow reload, but kills non-content parameters like “top_p” and “store” also. https://platform.openai.com/playground/prompts?models=gpt-4o

Good riddance prompts, a short-sighted, half-implemented, UI-driven UI destroyer.

Since Prompts is being shut down, please don’t leave the Playground stuck with the consequences of a dead API direction.

This is a good time to:

  1. Revert Prompt-specific Playground UI changes.
  2. Restore the simpler pre-Prompts editing flow where possible, especially stopping the promotion and redirection to Responses when using Chat Completions.
  3. Re-enable the Chat Completions presets editing and sharing that was purposefully broken for this sad promotion.
  4. Fix the editing, scrolling, and state-loss regressions.
  5. Remove terminology and workflows that only made sense for Prompts, versions, variables and other things of no use.
  6. Make the Playground reliable again for quick model testing.

Prompts was a bad fit, and the Playground should not remain shaped around it after shutdown. Please cleanly back it out instead of leaving users with the goblin leftovers - so bad one shares their own prompts (not the attempt to own language by overloading “prompts” as a name for a server-side settings) with their own Playground the way it should work. (Link might be too long for a onebox?)

MyPlayground - Goblin mode rewrites replies


June 25:

How about that … a “Chat” reorganization!

https://platform.openai.com/chat/edit?models=gpt-5.2

No longer intercepting attempts to use a playground with “make your prompt” screen.

It still has a fault that has never been fixed right since the “prompts->responses” change forced an “edit” button on you and forced a rendered markdown message view:

  • while editing a long message, you cannot use the overflow scroll bar to reach a different part of the message;
  • a click anywhere outside will automatically close the edit mode.
  • in the edit screen, the cursor caret moved to the bottom of such a long message by keyboard arrows will go off-screen, under the input box - you cannot see what you are newly typing.

Then:

  • Chat Completions presets are still locked up, locked down, unshareable (just like this topic’s first post), locked to shutoff models. A preset entry can actually be “responses”, completely switching the UI, then really messing up the flow of picking between the ones you made a year ago.