The short answer: OpenAI is deprecating its dataset-backed prompt optimizer along with the Evals platform. Existing evals became read-only on October 31, 2026, and the platform is scheduled to shut down on November 30, 2026 (OpenAI). If you ran datasets and graders, OpenAI's own deprecations page points to Promptfoo. If you only pasted prompts to improve them, a simple optimizer or a good meta-prompt does the job.
What's going away
| Item | What happens | When |
|---|---|---|
| Evals platform | Existing evals become read-only | October 31, 2026 |
| Evals dashboard and API | Scheduled to shut down | November 30, 2026 |
| Dataset-backed prompt optimizer | Deprecated as part of Evals | November 30, 2026 |
Reusable prompt objects and the v1/prompts API | Scheduled to shut down | November 30, 2026 |
The optimizer's guide also describes a chat view in the dashboard. Its dataset mode is the part tied to Evals by name. OpenAI's deprecations page doesn't list the optimizer as its own entry, so check your dashboard for how the chat view behaves.
What to use instead
You ran datasets and graders in OpenAI's dashboard. Export your evals now, since they're already read-only, and move to Promptfoo, the path OpenAI points to. It's open source and runs test cases across models. If you want a tool that tunes prompts for you from examples, look at DSPy, though it asks for more setup.
You stored prompts as reusable prompt objects. OpenAI's advice is to move that content into your application code. Do it before November 30, and keep the wording in version control so it can't vanish again.
You just wanted a better prompt. You don't need an evals platform. Your options:
- A meta-prompt. Ask any AI to improve your prompt, then test it. It's free, and usually good enough for a one-off.
- A paste-in optimizer. PromptSaver, our product, rewrites a prompt, shows its cost and names the cheapest good model at OpenAI, Anthropic and Google. See the other options too.
To be plain about what PromptSaver isn't: it doesn't take datasets or graders. It rewrites one prompt at a time, with an optional side-by-side check of the answers. If you need to score a prompt against hundreds of cases, use Promptfoo.
A safe way to move
- Export now. Download your eval runs, graders and datasets while you can still read them.
- Copy your prompts out, along with the model and settings each one ran on.
- Rebuild your best test set first. Start with 20 to 50 real cases, including the ones that have burned you before.
- Run the old prompt and any new prompt on the same cases, on the same model. OpenAI itself advises reviewing optimized prompts by hand before using them in production.
- Check the model while you're at it. Prices have moved a lot this year, and the AI prices page shows the current numbers. A smaller model often passes the same tests for much less.
Frequently asked questions
Is OpenAI's prompt optimizer being discontinued?
The dataset-backed version is, as part of the Evals platform, which is scheduled to shut down on November 30, 2026.
What does OpenAI recommend instead of Evals?
Its deprecations page points to a migration path to Promptfoo.
Will my prompts disappear?
Reusable prompt objects and the v1/prompts API are scheduled to shut down on November 30, 2026. OpenAI says to move that content into your application code, so copy it before then.
Does this affect ChatGPT?
These changes are to OpenAI's developer platform. A separate change does affect ChatGPT users: custom GPTs retire on December 11, 2026.
How we checked this
We read OpenAI's prompt optimizer guide and deprecations page on November 2, 2026. OpenAI can change its timelines, so follow the notices in your account. We'll update this page if the dates move. PromptSaver is our own product and isn't affiliated with OpenAI.