Michael JohnsonDevelopers who store model quality checks on the OpenAI Evals platform have been able to view existing evaluations, but not edit them, since October 31, 2026. OpenAI has scheduled the dashboard and API to shut down on November 30. These are separate stages in the transition: there is still time to save needed materials, but the platform is then expected to stop working. The change affects the hosted Evals service, not the practice of evaluating models itself—the timeline is listed in OpenAI’s product deprecation schedule.
What changed on October 31
Since October 31, existing evaluations have been available in read-only mode. The Evals dashboard and API are scheduled to shut down on November 30. The documentation describes a transition period, but does not promise automatic migration of user evaluations to another product.
While materials are still available, teams should save the data they need to repeat their checks: test examples, prompt instructions, expected answers, grader settings, and run results. This is a practical migration recommendation, not a list of guaranteed export formats: check what is available to your team and confirm that saved files can be opened outside the Evals interface.
What options does OpenAI offer instead of Evals?
The documentation suggests considering two scenarios. Datasets on the OpenAI platform are designed for rapid iteration: a team adds examples, runs prompt variants, evaluates results, and uses graders if needed. The documented types include exact string matching, text similarity scoring, model graders, and Python code execution.
For programmatic evaluation setup, the documentation describes the Evals API. However, according to OpenAI’s timeline, the API itself, like the dashboard, is also scheduled to shut down on November 30. It should therefore be viewed as a description of the previous workflow, not as a long-term replacement for the platform.
It’s best to start the migration with an inventory: what task did each dataset support—manual prompt comparison, an automated test in a release pipeline, or a batch run? OpenAI offers Datasets for rapid iteration on examples. For API-dependent and large-scale scenarios, the team will need to choose and test another workflow that supports the required operations and result storage.
What to save before the shutdown
Make a list of the materials needed to repeat evaluations: input examples, expected answers, prompt text, grader criteria and settings, and run results that matter for comparison. Then conduct a verification check using your chosen replacement: rerun a small set of tests and confirm that the evaluation criteria and interpretation of results remain comparable.
Reference data and evaluation rules help reproduce checks after a transition. If your team used external models, API runs, or large datasets, include those requirements in your criteria for choosing the next tool. OpenAI’s documentation describes Datasets and the former Evals API, but does not confirm that every previous scenario can be migrated automatically to Datasets.
Check which saving options are available to your account, and open the resulting files outside Evals before the scheduled shutdown. This test will show whether the data your team needs has been preserved, without relying on assumptions about export formats or automatic migration that the documentation does not promise.
Who is affected by the transition?
The timeline matters to developers who created or ran evaluations through the Evals dashboard, and to teams that depend on the Evals API. According to OpenAI’s schedule, both parts of the platform are due to shut down on November 30, 2026; the documentation does not say that the associated API will remain available after that date.
The practical next step is to save needed test materials and results, then check the new evaluation process on a small sample. This will help the team continue comparing models and prompts without depending on a platform that is closing.
