OpenAI Evals read-only November 30 shutdown is the deadline every OpenAI builder needs on the calendar this week. Existing evals became read-only on October 31, 2026, and the Evals dashboard and API shut down entirely on November 30, 2026.
November is the migration window OpenAI left on purpose: read your old evals, rebuild them somewhere you control, and compare results before the dashboard disappears.
Three shutdowns land on the same day
November 30 is a triple deadline. All three items were deprecated together on June 3, 2026, so teams have had months of warning — but the read-only date is what turns a plan into an emergency.
- Evals dashboard and API. Gone on November 30. Any evaluation suite running in continuous integration needs a new home before then, or the pipeline fails.
- Agent Builder. Gone the same day. Workflows built in it must move to the Agents SDK or ChatGPT Workspace Agents. ChatKit, the embeddable chat window, stays available.
- Reusable prompt objects and the v1/prompts API. Gone the same day. There is no replacement service: OpenAI’s advice is to move prompt content into your own application code.
Read-only means you can view existing evals but cannot run or change them. If a final benchmark run was still on your list, October 31 was the last day to do it.
What to save, in order
An evaluation is more than a list of questions: the valuable parts are its definitions and results.
First, pull grader definitions through the API: the data source configuration and testing criteria define the graders. A plain dashboard JSONL export holds less, so do not rely on it alone.
Second, pull every run and its output items: each row carries the input, the sample output, and each grader’s name, pass or fail, and score. Keep the raw JSON.
Third, download the dashboard JSONL for every run as a second copy. Then rebuild each eval in Promptfoo, OpenAI’s suggested migration tool, and compare results against the stored ones while the old dashboard still exists.
The multi-turn trap
Here is the part that catches teams out. A dataset eval — a table of inputs, outputs and grader verdicts — maps cleanly onto Promptfoo. A multi-turn test does not.
A multi-turn test is a scenario with a user, a goal and a stop rule, played out over several exchanges. Promptfoo’s migration guide covers dataset evals, but multi-turn scenarios have to be written again by hand, and conversation simulation needs its own setup.
Budget for that rewrite now: teams assuming “export and import” will discover in late November that their agent tests were never a table.
The evidence: dates and replacements
| Detail | What the source says |
|---|---|
| June 3, 2026 | Deprecation announced for Evals, Agent Builder and reusable prompts |
| October 31, 2026 | Existing Evals become read-only: no new runs, no edits |
| November 30, 2026 | Evals dashboard and API shut down |
| November 30, 2026 | Agent Builder shuts down; ChatKit remains available |
| November 30, 2026 | v1/prompts API and reusable prompt objects shut down |
| Evals replacement | Promptfoo, per OpenAI’s migration guide |
| Agent Builder replacement | Agents SDK or ChatGPT Workspace Agents |
| What to export | Grader definitions, run results and raw API JSON, plus dashboard JSONL |
| Known gap | Multi-turn scenarios must be rewritten, not imported |
Promptfoo is an open-source evaluation tool you run yourself. OpenAI’s own migration guide shows how to recreate a dataset eval in it.
Your November migration checklist, in order
1. Inventory — List every eval, Agent Builder workflow and stored prompt ID in your account.
2. Export via API — Pull grader definitions, runs and output items as raw JSON, plus dashboard JSONL.
3. Rebuild in Promptfoo — Recreate dataset evals first; rewrite multi-turn scenarios by hand.
4. Compare results — Run old and new side by side while the dashboard still exists.
5. Move prompts into code — Replace every stored prompt ID with text in your application.
6. Fix CI before Nov 30 — Point pipelines at the new evals so nothing goes red on shutdown day.
How to profit from it
Every shutdown creates a short, well-paid migration market. This one has a hard date, a clear scope, and thousands of teams that have not started.
- Sell a fixed-price Evals audit. Inventory a client’s evals, workflows and prompt IDs, and deliver a dated migration plan. Price it flat.
- Offer the Promptfoo rebuild. Rebuilding dataset evals is straightforward work with a clear deliverable: the same tests, running in a tool the client owns.
- Charge extra for multi-turn rewrites. Scenario-based agent tests are the hard part. Quote them separately, because they take real design work.
- Bundle the prompts migration. Moving stored prompts into application code touches every call site. Offer it as one package, not per prompt.
- Book November now. After November 30 the old results are gone, and a migration without a baseline to compare against is guesswork. Urgency is honest here.
Sources
OpenAI’s deprecations page as tracked by DEV Community and the migration analysis at AIVAX, both citing OpenAI’s deprecations page and migration guide. Re-check OpenAI’s deprecations page before scheduling around any date.
Frequently Asked Questions
What does read-only mean for OpenAI Evals?
Existing evals can still be viewed after October 31, 2026, but they cannot be run or edited.
When does the Evals dashboard disappear completely?
November 30, 2026, when the dashboard and API shut down.
What else shuts down on November 30?
Agent Builder, and the v1/prompts API with reusable prompt objects — all three on the same day.
Where should Evals move to?
Promptfoo, the open-source tool OpenAI points to in its migration guide.
Can I just export and import my evals?
For dataset evals, mostly yes. Multi-turn scenarios must be rewritten by hand, because they are not a simple table.
Keep Reading
Want more tools that actually pay for themselves? Browse the AI Tools section for the latest launches, pricing breakdowns, and profit playbooks — and check the official source linked above before you buy or deploy anything.