GPTMap

OpenAI Evals go read-only on 2026-10-31: the migration checklist before the 2026-11-30 shutdown

The deprecation page timeline (verified 2026-10-31): existing evals become read-only on 2026-10-31; the Evals dashboard and API shut down 2026-11-30. The same day also ends v1/prompts and Agent Builder. Official Evals path: Promptfoo.

TL;DR
The OpenAI deprecation page (verified 2026-10-31): existing evals become read-only on 2026-10-31 (announced 2026-06-03, 150 days); the Evals dashboard and API shut down 2026-11-30 (30 days later). Graders are part of the transition. Official path: Moving from OpenAI Evals to Promptfoo. The same calendar shuts down v1/prompts and Agent Builder on 2026-11-30; ChatKit remains available.
The Evals read-only milestone is the second execution date of OpenAI's 2026-06-03 deprecation batch: the official deprecation page marks Oct 31, 2026 as Existing evals become read-only, meaning existing evaluations on the Evals platform enter a read-only stage from that date, and the Evals dashboard and API shut down entirely on 2026-11-30. The same timeline also retires v1/prompts with reusable prompt objects and Agent Builder on 2026-11-30 — all three come from the same 2026-06-03 announcement.

The read-only milestone for OpenAI's Evals platform lands on 2026-10-31: the official deprecation page marks the date as Existing evals become read-only — existing evaluations on the platform enter a read-only stage from that day, ahead of the full dashboard and API shutdown on 2026-11-30. Based on the deprecation page as fetched on 2026-10-31, this article turns the batch — Evals, v1/prompts, and Agent Builder — into a checklist you can act on: the timeline, the boundaries, and the migration paths OpenAI actually names.

1. Which row takes effect today

The Evals section of the deprecation page (2026-06-03: Evals platform) is a three-row timeline:

  • June 3, 2026: Deprecation announced for the Evals platform.
  • Oct 31, 2026: Existing evals become read-only.
  • Nov 30, 2026: The Evals dashboard and API are scheduled to shut down.

Two boundary sentences right after the table define the scope of this batch: Graders documented for eval workflows are part of this transition — graders documented for eval workflows move with it; and Fine-tuning-related timelines remain covered in the self-serve fine-tuning section below — fine-tuning timelines do not follow this section and stay in the deprecation page's self-serve fine-tuning part.

An honest caveat: GPTMap holds no OpenAI API key, so on 2026-10-31 we could not verify item-by-item whether read-only mode is actually enforced inside the platform; on the day, the page still presented the row as a timeline entry. Treat the platform UI and official follow-ups as the source of truth for actual operability.

2. The timeline: 150 days and 30 days

The three rows, converted into intervals (calendar-command computed, not estimated):

MilestoneDateInterval
Deprecation announced2026-06-03—
Existing evals become read-only2026-10-31150 days after announcement
Evals dashboard and API shutdown2026-11-30180 days after announcement; 30 days after read-only

Those 30 days between read-only and shutdown are the migration window: beyond the three timeline rows, the page lists no additional buffer measures or exemptions — right after the table come the two boundary sentences (graders in transition, fine-tuning handled separately) and one migration pointer, and nothing else.

3. The other two rows on 2026-11-30: v1/prompts and Agent Builder

The same 2026-06-03 batch announced two more products with the same shutdown date:

v1/prompts and reusable prompt objects. Two timeline rows: June 3, 2026, Deprecation announced and prompt creation de-emphasized in the platform; November 30, 2026, The v1/prompts API and reusable prompt objects are scheduled to shut down. The official migration sentence: To migrate, move reusable prompt content into your application code — the guide page is Migrate from prompt objects.

Agent Builder. Two timeline rows: June 3, 2026, Deprecation announced for Agent Builder; November 30, 2026, Agent Builder is scheduled to shut down. One easily missed sentence in this section: ChatKit remains available — Agent Builder shutting down does not mean ChatKit shuts down. The official migration pointer is Migrate from Agent Builder, with the destinations being the Agents SDK or ChatGPT Workspace Agents.

Three products, two dates, one announcement — if any of them is in your dependency list, the migration window is the same one: through 2026-11-30.

4. Evals to Promptfoo: the officially named path

The Evals section's migration guidance is a single sentence: See Moving from OpenAI Evals to Promptfoo for a migration path. That is the destination OpenAI names for this batch — the cookbook guide Moving from OpenAI Evals to Promptfoo on developers.openai.com (the authors verified the page loads on 2026-10-31).

This article does not restate the guide's step-by-step instructions — for the actual migration procedure, follow the cookbook page. What this article confirms for you: the path is the one the deprecation page itself names, the page loads today, and it closes the loop with this batch's timeline (read-only 10-31, shutdown 11-30) inside the same section.

5. A checklist for the last 30 days

Based on the page's three timeline rows and two migration guides, worth confirming one by one before 11-30:

  1. Evals: do your workflows still read evaluation configs from the Evals platform? If so, follow Moving from OpenAI Evals to Promptfoo; graders used by those workflows are inside this transition too — do not migrate the evals and leave the graders behind.
  2. v1/prompts: any remaining v1/prompts calls or reusable prompt object references in your code? The official path is moving the content into application code (Migrate from prompt objects).
  3. Agent Builder: any agent flows still built in Agent Builder? The official destinations are the Agents SDK or ChatGPT Workspace Agents (Migrate from Agent Builder). If your chain only uses ChatKit, note the page verbatim — ChatKit remains available — so first clarify whether you depend on the Builder or the Kit.
  4. Fine-tuning: if your Evals usage ties into fine-tuning evaluation, remember the page states fine-tuning timelines live in the self-serve fine-tuning section — a separate set of dates; do not apply this article's 11-30 to fine-tuning assets.

Every item above anchors to the deprecation page text; where the page is silent (export formats, transitional API behavior), this article does not guess.

6. Where this sits in the larger shutdown calendar

This batch is not an isolated event. From the same deprecation page, verified 2026-10-31, the surrounding rows are:

  • 2026-12-01: the older GPT Image models row (announced 2026-06-02) — gpt-image-1-mini and siblings, replaced by gpt-image-2.5-sunburst or the flare variant.
  • 2026-12-11: the GPT-5 and o3 snapshots row (announced 2026-06-11) — gpt-5-2025-08-07 and siblings, replaced by gpt-5.6-sol and matching tiers.
  • 2027-01-14: the GPT-4 series rows (announced 2026-10-14) — gpt-4o-2024-08-06, gpt-4o-2024-11-20, and siblings, replaced by gpt-6-luna.

For quarterly migration planning, 11-30 (these three products) and 12-11 (GPT-5/o3 snapshots) are the next two hard dates; the row-by-row official replacement table lives in the shutdown-day article linked below.

7. Boundaries and common misreadings

  • Read-only is not shutdown: 10-31 changes writability only; no row on the page says evaluations are deleted on 10-31. The deletion window is the 11-30 dashboard and API shutdown.
  • ChatKit does not go down with Agent Builder: that is page verbatim (ChatKit remains available), not this article's inference.
  • Graders are in scope, fine-tuning is not: both sentences come from the Evals section text; folding fine-tuning timelines into this batch is a common misreading.
  • Verification boundary: the official page was fetched on 2026-10-31; whether read-only is enforced item-by-item in the platform, and how well third-party tooling supports the Promptfoo migration, are beyond the page text and this article makes no claims there.

8. Next steps

Key points

  • Verified against the official deprecation page on 2026-10-31: the Evals section timeline reads Oct 31, 2026 — Existing evals become read-only; GPTMap holds no API key, so we could not test on-platform read-only behavior on the day — actual availability follows the platform and official follow-ups
  • All three timeline dates come from the page verbatim: announced 2026-06-03; existing evals read-only 2026-10-31; Evals dashboard and API shutdown 2026-11-30 — 150 days from announcement to read-only, 180 days to shutdown, 30 days from read-only to shutdown (calendar-command computed)
  • Graders and fine-tuning boundary (page verbatim): Graders documented for eval workflows are part of this transition; Fine-tuning-related timelines remain covered in the self-serve fine-tuning section below
  • Official migration pointer: Moving from OpenAI Evals to Promptfoo (cookbook page, loaded on 2026-10-31) — Promptfoo is the migration path the deprecation page names for this batch
  • Two more rows shut down on the same 2026-11-30 date: v1/prompts and reusable prompt objects (official guide Migrate from prompt objects — move reusable prompt content into your application code; the same 2026-06-03 row also notes prompt creation de-emphasized in the platform); Agent Builder (official guide Migrate from Agent Builder — continue with the Agents SDK or ChatGPT Workspace Agents, with the batch verbatim ChatKit remains available)
  • Source phrasing: The v1/prompts API and reusable prompt objects are scheduled to shut down; Agent Builder is scheduled to shut down — both are main-table shutdown rows, not previews

Frequently asked questions

The deprecation page says exactly one line: Existing evals become read-only. It does not enumerate which operations are blocked or still allowed after that; GPTMap holds no API key to test platform behavior. Before the 2026-11-30 shutdown, treat the platform UI as the source of truth for what remains operable, and complete the migration following the official guide as soon as possible.

Official references

Related articles

Subscribe to GPTMap Weekly

One email every Monday: curated OpenAI updates, deep dives, and best practices. No ads, unsubscribe anytime.

Submitting opens Buttondown in a new tab to confirm your subscription.

GPTMap EditorialPublished 2026-10-31 6 min read
Test environment (EEAT)
Last tested: 2026-10-31
Model used: Platform products: Evals / v1/prompts / Agent Builder (model-agnostic); replacement ecosystem: Agents SDK / ChatGPT Workspace Agents / Promptfoo