API
OpenAI
OpenAI is retiring its hosted Evals platform: existing evals become read-only on 31 October 2026, and the Evals dashboard and API shut down entirely on 30 November 2026
departs 30 Nov 2026
Quick answers
- Is it true that OpenAI shuts down the Evals dashboard and API on 30 November 2026 (evals go read-only 31 October)?
- Yes. OpenAI has officially announced it, effective 30 Nov 2026. OpenAI's [deprecations page](https://developers.openai.com/api/docs/deprecations) lists the hosted **Evals** platform — the dashboard and the Evals API used to define, run and store model evaluations — on a two-step wind-down: **31 October 2026** existing evals become **read-only** (you can still view results but can no longer create or run new ones through the platform), and **30 November 2026** the **Evals dashboard and API shut down** entirely.
- When does it take effect?
- 30 Nov 2026. Verified against the vendor's own announcement.
- What should I use instead?
- Export anything you need out of the Evals dashboard **before 31 October 2026**, while it is still fully writable, and in all cases before 30 November when the data goes away with the platform — eval definitions, graders, datasets and the run history you want to keep.
What this means for you
OpenAI's [deprecations page](https://developers.openai.com/api/docs/deprecations) lists the hosted **Evals** platform — the dashboard and the Evals API used to define, run and store model evaluations — on a two-step wind-down: **31 October 2026** existing evals become **read-only** (you can still view results but can no longer create or run new ones through the platform), and **30 November 2026** the **Evals dashboard and API shut down** entirely. The deprecation was first announced on 3 June 2026. If you built an evaluation pipeline on OpenAI's hosted Evals — CI steps that call the Evals API to grade model output, graders and datasets stored in the dashboard, regression suites you run before shipping a prompt change — those API calls stop resolving on 30 November, and the eval definitions and run history living in the dashboard become inaccessible once it is gone. The read-only cutover on 31 October is the earlier trap: automation that creates or runs evals (not just reads them) breaks a month before the full shutdown. This does not touch the chat/completions API or any model — only the Evals product.
Where to go
Export anything you need out of the Evals dashboard **before 31 October 2026**, while it is still fully writable, and in all cases before 30 November when the data goes away with the platform — eval definitions, graders, datasets and the run history you want to keep. Then move your evaluation workflow off the hosted platform: OpenAI's open-source [evals framework](https://github.com/openai/evals) runs the same style of evals in your own CI, and third-party harnesses (promptfoo, Braintrust, LangSmith and similar) cover the dashboard-and-grading workflow. Repoint any CI step or script that calls the Evals API to your chosen replacement before the 30 November shutdown. OpenAI's deprecations page sets the dates but does not name a one-to-one hosted successor, so treat this as a migration to a self-hosted or third-party eval stack rather than a drop-in swap.
- openai/evals — OpenAI's open-source evaluation framework, run in your own CI
- Third-party eval platforms (promptfoo, Braintrust, LangSmith) for the dashboard + grading workflow
OpenAI’s API deprecations page schedules the hosted Evals platform — the dashboard and Evals API for defining, running and storing model evaluations — for shutdown in two steps:
- 31 October 2026 — existing evals become read-only. You can still view them, but you can no longer create or run new evals through the platform.
- 30 November 2026 — the Evals dashboard and API shut down entirely.
The deprecation was first announced on 3 June 2026. This affects only the Evals product; the chat/completions API and the models themselves are untouched.
Where this hurts
If you only ever opened the Evals dashboard by hand, this is a prompt to export your data and move on. It bites hardest where Evals is wired into automation: a CI step that calls the Evals API to grade a model’s output before a prompt change ships, a nightly regression suite, graders and datasets that live only in the dashboard. Two things break, on two different dates:
- The read-only cutover on 31 October stops any automation that creates or runs evals — a month before the full shutdown, and easy to miss if you were only watching the 30 November date.
- The 30 November shutdown stops the Evals API calls outright and takes the dashboard — and any eval definitions and run history stored only there — with it.
What to do
Export everything you need out of the dashboard before 31 October, while it is still fully writable: eval definitions, graders, datasets, and whatever run history you want to keep. Then move the workflow off the hosted platform. OpenAI’s own open-source openai/evals framework runs the same kind of evals in your own CI; third-party harnesses such as promptfoo, Braintrust and LangSmith cover the dashboard-and-grading workflow. Repoint any CI step or script that calls the Evals API before 30 November.
OpenAI’s deprecations page sets the dates but does not name a one-to-one hosted successor, so plan for a migration to a self-hosted or third-party eval stack rather than a drop-in replacement. This is separate from — but on the same page as — the large model removals on 23 October and the GPT-5 snapshot, o3 and o3-pro removals on 11 December; anything that touches the OpenAI platform is worth a periodic check against that page.
Entry changelog
- — last reviewed.
- — Entry created from OpenAI's official API deprecations page, which lists the hosted Evals platform with two dates: existing evals read-only 31 October 2026, dashboard and API shut down 30 November 2026 (deprecation announced 3 June 2026). No detect fingerprint: the Evals surface is reached through the standard OpenAI SDK/REST client rather than a distinct package or pinned model id, so any string specific enough to match it (a bare `evals` path) would also false-fire on unrelated code — omitted rather than guessed (SOURCING step 6).
Put this countdown in your README
Building on OpenAI? Drop this badge into a project’s README or docs — it shows the days left and
refreshes on its own every night from this page. 
[](https://lastcall.dev/entries/openai-evals-platform-shutdown)Spot an error or a changed date? Tell us and it’s corrected the same night.