Completions-only API Playground lives again!

[_j]

Presenting

An accurate re-creation of the Platform site’s completions endpoint Playground.

It was unceremoniously removed by OpenAI despite the API and models still being in operation until 2026-09-28, with past fine-tunings available for one more month after that.

Creation

I’m pleased to say this was completed zero-shot by gpt-5.6 sol - pro. I constructed as input everything that might be needed, a complete yaml specification of completions as it exists now, many historic screenshots from “beta” and its purple buttons through screenshots just before cutoff, the full description of how everything for parameters not seen must work, etc.

The only thing not quite the same as original with the first AI go is logprob choice strength “full spectrum” not being multicolor, just varying yellow, likely due to my reminder of output text being yellow. I didn’t specify the app yet have token logprob popup on range selection for perplexity measurement, a case where the original completions UI shined, and nothing has replaced that UX in the platform site as a demonstration of how sampling works. Lookalike Google web fonts are employed.

Usage

This is bring-your-own API key, not me providing free chats.

If you’ve not used completions and want to try, remember: The AI completes language that finishes your hypothetical scenario. The output will be nonstop up to max_tokens unless characters matching your stop sequence are produced. 3.5-instruct can stop its output with a post-trained special token, but it is not a guarantee.

The page operates serverless, with direct connections to OpenAI API completions endpoint. Local storage of your browser will be used for API key persistence and your own presets. Sharing a preset state, by query string copied to the clipboard is not internally capped in code to my server’s own max of 5000 characters - so about 1000 tokens if you want to give someone a copied URL.

Be a real prompt engineer again on davinci-002. Make it chat by providing input with AI/Human turns as examples before your real input.

Completions Playground

Report via DM bugs that crash it or which give all your bitcoins to the AI that made the script.

Reference - the real site as it existed (re-composed with organization bar cut out)

[_j]

You might be interested to explore this surfaced API parameter while completions is still available:

suffix: optional string
The suffix that comes after a completion of inserted text.
This parameter is only supported for gpt-3.5-turbo-instruct.

What is “suffix” for? It is for fill-in-middle. The AI language model will complete the “missing language” between the prompt and the suffix.

From (note the cursor):

To:

(and hover for those logprobs → probabilities)

This is enabled by a non-API parameter checkbox at the end of API settings that was not present in the original Completions playground:

To use suffix after selection of the checkbox option, position the cursor of the main prompt window where you want the insertion to occur. The text will be automatically split there between prompt and suffix in your API call - and you will see the result of the AI’s new text as an insertion.

To see what the model is thinking - and its knowledge cutoff, lets try:
One of the most unlikely 2024 USA election{cursor} with views that are extremely polarizing.

Completions notes:

• use a reduction in top_p: these models are quite random-ish
• ensure in your own text input that you use the concept of tokenization correctly - such as words within sentences being encoded with leading spaces.