TellWorldAI World Archive
Ask中文

Home›Practice›Practice

Practice

Migrating to GPT-6.2: prefix edits, instructions and prompting

Migrating to GPT-6.2 means checking three things: edits to earlier conversation history drop reasoning items, top-level instructions now sit after input, and several sampling parameters are no longer accepted.

Who this is for and what you will get

If your application already calls OpenAI models through the API and you plan to move it to GPT-6.2, this guide covers the changes that can quietly affect answer quality, token use and caching. After reading it you should know which parts of your request code to check, which parameters to remove, and how to write instructions for the new models.

What GPT-6.2 offers

OpenAI names two models in this family. GPT-6.2 Astra is meant for the most demanding tasks, and GPT-6.2 Sol is meant for complex work at a lower cost.[1] The API model identifiers are gpt-6.2-astra and gpt-6.2-sol.[1]

Prefix edits drop reasoning items

Starting with GPT-6.2, the Responses API drops reasoning items after mid-conversation edits to the conversation history. OpenAI calls these prefix edits.[1] The same thing happens after certain model changes, including downgrades, and after changes to the end user or the organization.[1]

When the model loses its reasoning items, answer quality can decline, and the model may spend more tokens repeating work it already did.[1] Look in your integration for edits such as changes to an earlier developer message or to an earlier tool call.[1] To keep reasoning and keep earlier input eligible for prompt caching, leave the prefix unchanged when you change settings between responses.[1]

Where instructions go

Models starting with GPT-6.2 append the top-level instructions field after input in the Responses API.[1] Instructions you want cached should go into a developer message placed first in input. Top-level instructions fit guidance that applies only to the current request.[1]

Parameters to change

  • Model: set model to gpt-6.2-astra or gpt-6.2-sol.[1]
  • Reasoning effort: for a new integration, start with medium. If you use none or minimal today, switch to low. Use reasoning.effort in Responses or reasoning_effort in Chat Completions.[1]
  • Tool calling: use the Responses API. Both models accept Chat Completions requests, but tool calling requires Responses.[1]
  • Unsupported parameters: remove temperature, top_p, logprobs and top_logprobs from Chat Completions requests. In Responses requests, remove temperature, top_p and top_logprobs, and take message.output_text.logprobs out of include.[1]
  • Caching: when migrating from GPT-5.5 or earlier, replace prompt_cache_retention with prompt_cache_options.ttl, set to 30m.[1] Review the caching changes as well, including cache boundaries and how cache writes are billed.[1]

Zero Data Retention and multi-turn state

Approved organizations can use Zero Data Retention (ZDR) with both models on eligible endpoints and features. With ZDR enabled, the Responses API treats store as false.[1] For multi-turn work, pass the earlier output items back in their original order, including the encrypted reasoning items.[1]

Data residency and service tiers

Both models support US and EU data residency, with Standard, Fast and Batch processing. Ultrafast supports US data residency only.[1]

How to prompt GPT-6.2

Astra writes more concisely and directly by default. If your system prompt says “be concise”, review that line when you migrate.[1] When the user's intent is unclear, the model may propose a next step and wait for confirmation. For longer tasks, tell it to make reasonable assumptions, keep working and state those assumptions.[1]

For long agent runs, OpenAI suggests stating that the final answer must stand on its own, because users may never read the earlier progress messages.[1]

For coding work, describe the conventions the codebase follows, including when to reuse existing logic and when to add tests. OpenAI's sample prompt asks the model to reuse existing logic in the user's repository and to skip new tests unless they are requested.[1]

For frontend work, if the interface needs a particular look, name the imagery and typography it should use.[1]

Next steps

To migrate an API application with a coding agent, OpenAI points to the OpenAI Docs skill in the openai/codex repository.[1]

Sources

  1. 1
    Using GPT-6.2OpenAI API docs

Cite this page: Migrating to GPT-6.2: prefix edits, instructions and prompting. TellWorld, https://www.tellworld.net/practice/practice-gpt-6-2-migration. Updated 2026-11-18

Related news

View all