OpenAI o3 reached its announced retirement date in ChatGPT on August 26, 2026. OpenAI first published the 90-day sunset notice on May 28 and explicitly limited the change to ChatGPT: the company said there was no change to API access.

For QA teams, this is less about losing a model name and more about controlling a test-environment change. Saved prompts, Custom GPT workflows, review checklists, and human-in-the-loop test processes can behave differently when the selected model changes. Teams should record the cutoff date and recheck the workflows that depended on o3 inside ChatGPT.

What changed on August 26

  • Product surface: OpenAI o3 was scheduled to retire from ChatGPT.
  • Effective date: August 26, 2026, after a 90-day sunset period.
  • Scope: OpenAI’s notice applies to ChatGPT, including the paid-user model selection context described in the release note.
  • API status: OpenAI explicitly said the announcement did not change API access.

That scope matters. A tester seeing o3 disappear from the ChatGPT model picker should not conclude that an API-based evaluation harness has also stopped working. ChatGPT and the API are separate test surfaces with separate lifecycle checks.

Why this matters for QA engineers

Model selection is part of the environment for any repeatable AI-assisted QA task. If a team used o3 to generate boundary cases, critique Playwright tests, summarize CI failures, or convert logs into defect reports, a replacement model may change output structure, prioritization, and the amount of human review required.

  • Golden outputs may drift: expected sections, severity labels, or edge-case lists may change even when the prompt and input stay fixed.
  • Prompt instructions may need tuning: formatting constraints that worked with o3 may be interpreted differently by a newer model.
  • Runbooks can become misleading: screenshots and team instructions that tell testers to select o3 are now stale.
  • Audit trails need context: August 26 should be marked as an environment-change date when comparing before-and-after results.

A focused regression check

Do not rerun every AI-assisted workflow at once. Start with the two or three prompts that influence the most consequential QA decisions.

ChatGPT model-transition check
1. Choose a saved QA prompt previously run with o3.
2. Reuse the same test artifact or failure log.
3. Compare coverage, factual errors, format, and actionability.
4. Record the current model and the August 26 transition date.
5. Update the prompt or review checklist if quality moved.

Pay special attention to prompts used for release decisions, security review, accessibility findings, or defect severity. Those workflows deserve stronger validation than low-risk brainstorming tasks.

What QA teams should do now

  • Replace o3-specific ChatGPT instructions in team runbooks.
  • Rerun a small set of known prompts and compare results with stored examples.
  • Document the active ChatGPT model whenever prompt output becomes test evidence.
  • Keep API monitoring separate; this retirement notice did not announce an API change.

The broader lesson is simple: model retirements should be handled like browser, dependency, or test-data changes. Name the environment change, test the critical path, and preserve enough metadata to explain output drift later.

Bottom line

The ChatGPT o3 retirement took effect on August 26, 2026, following OpenAI’s announced 90-day sunset. QA teams that used o3 through ChatGPT should revalidate their highest-value prompts and update documentation. API users should not infer an API retirement from this product-specific change.

Sources