OpenAI retired GPT-5.5 from ChatGPT, ChatGPT Work, and Codex on October 14, 2026. The company’s model release notes make one scope boundary especially important: the retirement does not affect API access. For QA engineers, this is a practical environment change wherever repeatable review, test-design, or coding workflows depended on GPT-5.5 in those products.

This is not a claim that every workflow will regress. It is a reason to recheck the few workflows where model behavior is part of the expected outcome: a saved defect-triage prompt, an assistant-led test-plan review, or a Codex task that follows team instructions and produces evidence for review.

What OpenAI confirmed

  • Retired model: GPT-5.5.
  • Effective date: October 14, 2026.
  • Affected products: ChatGPT, ChatGPT Work, and Codex.
  • Not affected: API access, according to OpenAI’s release note.

That product boundary matters for incident triage. A team should not infer that an API evaluation harness must change simply because a ChatGPT or Codex workflow changed. Record the product surface, model, date, and user-visible behavior separately.

Why this matters for QA engineers

AI-assisted QA is often partly human-in-the-loop, but that does not make it untestable. When a model is retired, wording, structure, prioritization, tool choices, and the amount of reviewer follow-up can shift. Treat the change as a controlled baseline update—not as proof of a defect.

  • Prompt outputs can drift: the same bug report may produce different edge cases or severity reasoning.
  • Codex review habits can change: task decomposition, test suggestions, and summaries may need a fresh sample review.
  • Runbooks can go stale: screenshots and instructions that name GPT-5.5 may no longer match the available product.
  • False incidents become less likely: marking October 14 as an environment-change date gives teams context when comparing historical results.

A focused regression check

Model-transition spot check
1. Pick one high-value QA prompt or Codex task.
2. Reuse the same input artifact or repository fixture.
3. Compare output format, factual grounding, coverage, and actionability.
4. Record the current model/product surface and October 14, 2026 change date.
5. Update team guidance only if the new result needs it.

A small, repeatable check is more useful than a broad unstructured retest. Good candidates are a flaky-test diagnosis, a requirements-to-test-cases prompt, or a narrowly scoped Codex change with an existing expected review checklist.

Keep the API and product checks separate

OpenAI explicitly says this retirement does not affect API access. Keep that statement intact in tickets and release notes: validate ChatGPT, ChatGPT Work, and Codex workflows where they are used, while leaving an API integration untouched unless independent evidence calls for a change. This avoids unnecessary churn and makes investigations easier to audit.

Bottom line

The OpenAI GPT-5.5 retirement is a concise but meaningful QA signal. GPT-5.5 is retired from ChatGPT, ChatGPT Work, and Codex; API access is unchanged. Re-run a handful of trusted model-dependent workflows, document the transition date, and keep product-surface testing separate from API validation.

Source