OpenAI Evals goes read-only on 31 October; Agent Builder ends 30 November

OpenAI Evals goes read-only on 31 October; Agent Builder ends 30 November

23 October 2026: Three OpenAI developer products reach their end in the next six weeks. Existing evals on the OpenAI Evals platform became read-only on 31 October 2026, and on 30 November the Evals dashboard and API, the v1/prompts API with its reusable prompt objects, and Agent Builder are all scheduled to shut down. OpenAI announced all three on 3 June; teams that have not migrated have until 30 November.

Not yet independently verified. The dates are those on OpenAI's deprecations page, which gives 3 June 2026 as the notice date and was re-read on 10 November 2026. The 31 October read-only step has passed and the 30 November shutdowns are still ahead. We will update this when it can be confirmed, and remove this note.

Update, 10 November 2026: OpenAI's deprecations page, re-read today, prints 3 June 2026 as the notice date for all of these changes. It still lists existing evals as read-only from 31 October 2026, and the Evals dashboard and API, Agent Builder and the reusable prompts (v1/prompts) API as shutting down on 30 November 2026. The read-only step has now passed.

What closes and where OpenAI points users

Date	What changes	OpenAI's suggested route
31 October 2026	Existing evals become read-only	export results and move to another eval tool
30 November 2026	Evals dashboard and API shut down	Promptfoo, per OpenAI's migration guide
30 November 2026	v1/prompts API and reusable prompt objects shut down	move prompt content into application code
30 November 2026	Agent Builder shuts down	Agents SDK or ChatGPT Workspace Agents; ChatKit stays
1 December 2026	gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest removed	gpt-image-2.5-sunburst or gpt-image-2.5-flare

The graders documented for eval workflows are part of the Evals transition, OpenAI says, while fine-tuning has its own timeline.

Why the 31 October date matters more than it looks

Read-only evals can still be viewed but not run, so any regression test suite that calls the Evals API as part of a release pipeline stops being useful a month before the shutdown itself. Teams that gate model or prompt changes on OpenAI-hosted evals should move those tests, and the datasets and graders behind them, before 31 October, not 30 November. Reusable prompt objects carry a similar hidden dependency: applications that fetch a prompt by ID at run time will fail when the endpoint goes, so the prompt text has to be copied into code or a prompt store and redeployed first.

Today's model shutdowns and what follows

The deadlines sit inside a longer retirement schedule. On 23 October OpenAI's page lists the shutdown of a set of legacy snapshots it announced in April, including gpt-3.5-turbo, gpt-4, gpt-4-turbo, gpt-4o-2024-05-13, gpt-image-1, o1, o1-pro and o3-mini, which we covered in September. Further dates on the same page include 11 December for a group of GPT-5 and o3 snapshots, 6 January 2027 for new fine-tuning jobs by existing customers and for the older text-to-speech models, and 14 January 2027 for GPT-4o and GPT-4.1, which we reported on 14 October.

Sources

• OpenAI API documentation: Deprecations (read 2026-10-23), 23 October 2026
• OpenAI API documentation: Deprecations (re-read 10 November 2026), 10 November 2026