OpenAI Evals Platform and Prompt Objects Shutting Down November 2026: Migration Guide

On June 3, 2026, OpenAI announced the deprecation of two platform features: the Evals platform and reusable prompt objects. Both services share the same shutdown date — November 30, 2026 — giving builders roughly five months to migrate.

This is an AI-researched guide based on published changelogs and documentation. ChatForest does not have hands-on access to these tools.

What is being shut down

Evals platform

The Evals platform let you define test datasets, write graders, and run automated eval jobs against model outputs. Builders used it to validate that prompts and model responses met style, accuracy, or content criteria before shipping changes.

Timeline:

• June 3, 2026 — Deprecation announced; no new eval infrastructure added
• October 31, 2026 — Platform becomes read-only; existing evals can no longer be run or edited
• November 30, 2026 — Full shutdown; all eval data and runs deleted

Reusable prompt objects

Prompt objects were managed templates stored in OpenAI’s platform, referenceable by ID and version with dynamic variable substitution. The API accepted a prompt parameter pointing to a stored object instead of requiring the prompt text inline.

Timeline:

• June 3, 2026 — New prompt creation de-emphasized in the dashboard
• November 30, 2026 — v1/prompts endpoint shuts down; all stored objects deleted

Agent Builder (covered separately)

Agent Builder was also deprecated on June 3, 2026, with the same November 30 shutdown date. The migration path is the Agents SDK or ChatGPT Workspace Agents. See the OpenAI Agent Builder and Assistants API deprecation guide.

Migrating off the Evals platform

OpenAI’s suggested migration path is Datasets — an iterative environment for experimenting with eval workflows as you build.

For teams that want a dedicated external eval harness, Promptfoo is OpenAI’s cited third-party alternative. Promptfoo is open-source, supports OpenAI models through the standard API, and runs eval suites in CI pipelines.

Migration steps for Evals:

1. Export or document your existing eval configurations while the platform is still writable (before October 31).
2. Download any datasets you want to retain — the platform provides no export after read-only mode.
3. Recreate eval logic in Datasets or in Promptfoo config files.
4. Wire Promptfoo (or your chosen tool) into your CI pipeline where OpenAI Evals previously ran.

Do this before October 31 — after that date you can read existing runs but cannot export, edit, or re-run evals.

Migrating off prompt objects

The migration philosophy here is straightforward: move prompt content out of OpenAI’s managed store and into your application code, where it can be versioned in git and reviewed in pull requests like any other product logic.

Migration steps:

1. Extract the prompt text. Open each prompt object in the dashboard and copy the full text into a constant or file in your codebase.

2. Replace variables with function arguments. Prompt objects supported template variables (e.g., {{user_query}}). Convert these to typed function parameters in your code.

3. Update API calls. If you used the prompt parameter in Responses API calls, replace it with the input field containing your messages directly:

   # Before (prompt object)
   response = client.responses.create(
       model="gpt-5-5",
       prompt={"id": "prompt_abc123", "variables": {"user_query": query}}
   )
   
   # After (inline messages)
   response = client.responses.create(
       model="gpt-5-5",
       input=[
           {"role": "system", "content": SYSTEM_PROMPT},
           {"role": "user", "content": query}
       ]
   )
   

4. Implement versioning in code. Use git commits and your existing CI/CD pipeline for prompt change management. This is why OpenAI is making this change: prompt changes treated as code changes get review, testing, and rollback.

Why OpenAI is making these changes

For prompt objects: moving prompt management into application code gives teams “more control over review, testing, deployment, and versioning.” Prompts are product logic; they belong in git, not in a separate managed store that bypasses normal code review.

For Evals: the platform served a specific testing workflow that OpenAI is now simplifying. The Datasets interface and third-party tools like Promptfoo provide more flexibility for the range of evaluation patterns teams actually use.

Action calendar

The five-month window is enough time, but October 31 is the functional deadline: after that you cannot run evals or export data, only read existing results.