Legacy Model Workflow Migration Checklist: Retest Fixed Inputs and Acceptance Criteria

This URL was first published as a GPT-5 usage guide on August 9, 2025. GPT-5 Instant and Thinking were retired from ChatGPT on February 13, 2026, so old picker screenshots and “hidden tips” are not current instructions. This correction explains how to migrate the underlying workflow.

This legacy model workflow migration checklist retests fixed inputs against explicit acceptance criteria.

Separate ChatGPT from the API

OpenAI's retirement notice describes ChatGPT model changes and says the retirement did not change the API at that time.

• ChatGPT: check the live selector, tools, account conditions, and release notes.
• API: check current models, deprecations, pricing, and parameters in API documentation.

The same GPT-5 label does not establish the same availability across the product and developer platform.

Migration record

Surface: ChatGPT / API
Feature dependency:
Evaluation input:
Acceptance criteria:
Material difference from the old result:

ChatGPT conversations

Continuing an old conversation may use a newer replacement model and produce different output. Retest important templates in a new conversation using the same input. Review numbers, citations, format constraints, and prohibited actions.

API code

Do not replace only the model string and deploy. In staging, verify:

1. request format and supported parameters;
2. structured output and tool-call compatibility;
3. pass rate on known evaluation examples;
4. latency, token usage, and current price;
5. fallback and stop behavior.

Do not use prices or model composition in the 2025 developer announcement as an undated current value.

Current sources of truth

• launch, update, and retirement: Model Release Notes;
• ChatGPT options: the account performing the work;
• API: current official model, price, and deprecation documentation;
• factual output: the primary source opened by the reviewer.

Conclusion

An old GPT-5 guide is not a current ChatGPT operating manual. Record the surface, dependency, evaluation input, acceptance criteria, and differences, then migrate ChatGPT and the API separately. Preserving the original URL and publication date makes the correction visible instead of hiding the article's history.