Choosing Instant, Auto, or Thinking With a Current Task-Routing Matrix
Replace the retired GPT-5 Fast, Auto, and Thinking table with a routing matrix based on constraints, verification, time, and failure impact.
2 min read
This article originally compared GPT-5 Fast, Auto, and Thinking. GPT-5 Instant and Thinking were retired from ChatGPT in February 2026. Do not use that old feature table as a current specification. Use criteria that apply to whatever normal, automatic, or reasoning options the current account provides.
Classify the work on four axes
| Axis | Low | High |
|---|---|---|
| Constraints | summary or format conversion | multi-constraint plan or design |
| Verification | immediate source comparison | tests or several sources required |
| Latency tolerance | rapid iteration matters | a slower single pass is acceptable |
| Failure impact | disposable draft | publication, contract, or safety impact |
A high-impact task is not made safe by selecting a reasoning option. Add primary sources, tests, and accountable approval.
Routing guidance
Normal or Instant-like
Start here for source-grounded summaries, format conversions, and idea candidates that can be checked and retried quickly.
Auto-like
This may fit mixed-complexity work when an automatic switch is present in the current ChatGPT interface. Because a user may not always observe every routing detail, score the result against acceptance criteria rather than trusting a model label.
Thinking-like
Try this for multi-constraint design, comparison, or code verification when a small controlled test shows a meaningful benefit. Record both the result and added time.
Example
Start a 300-word source summary with the normal option and compare it with the source. For three proposals constrained by a law, budget, and deadline, compare a reasoning option—but verify the legal rule in an official source and retain qualified human judgment.
Confirm current behavior
Use OpenAI's retirement notice, Model Release Notes, and the live account interface. Availability can vary by plan and workspace.
Conclusion
Do not memorize retired GPT-5 mode names. Route by constraints, verification, time, and failure impact. Start with the normal option, compare automatic or reasoning choices only when useful, and retain human review for consequential work.
Primary sources checked
Important claims should also link to the relevant source in the article body.
- Retiring GPT-4o and other ChatGPT modelsOpenAI Help Center · official-help · Checked: 2026-07-26
- Does ChatGPT tell the truth?OpenAI Help Center · official-help · Checked: 2026-07-26
- Model Release NotesOpenAI Help Center · official-release-note · Checked: 2026-07-26