Before Continuing an Old GPT-5 Chat: Retest After a Model Change

This page originally compared GPT-5 Instant and Thinking. Both were retired from ChatGPT on February 13, 2026. OpenAI explains that old conversations can continue on replacement models and that their output may differ.

Preserve five fields before continuing

Do not preserve only the final answer. Preserve the conditions and review method.

Run a fixed test in a new conversation

1. Select three representative inputs without personal or confidential data.
2. Submit the same instructions in a current new conversation.
3. Compare old and current results line by line.
4. Score numbers, citations, format, prohibited actions, and tool behavior.
5. Stop production use if the result misses its acceptance criteria.

A long conversation may also contain conflicting old conditions. A new conversation tests that separately from the model replacement.

Recheck these outputs

• current laws, prices, specifications, medical, and safety information;
• generated URLs and citations;
• code, formulas, and structured data;
• external changes to email, calendars, or files;
• claims that rely only on what the conversation previously said.

OpenAI states that ChatGPT can generate incorrect facts and fabricated sources.

Confirm the current model

Check the retirement notice, Model Release Notes, and the options shown in the actual account.

Conclusion

Before continuing a retired GPT-5 conversation, preserve assumptions, format, citations, actions, and decisions, then run fixed tests in a current new conversation. The ability to open an old chat does not guarantee the same model or output.