How to use multiple models in a Codex workflow

You can change the model between turns using /model.

There are two ways to approach this:

1. Start with gpt-5.4-mini, then switch to gpt-5.5 in the next turn. Note that this will send all previous turns in the thread to gpt-5.5 as uncached context.

2. The preferred method, if your goal is to save the most usage limits:

   • Use gpt-5.4-mini to create a file, such as report.md, containing the results.
     • Prompt: Please analyze <your giant task> and save the report in report.md.
   • Start a new thread with /new, switch to gpt-5.5, and use @report.md to refer to the file. This discards all previous reasoning and initial inputs, using only the summarized report instead, which saves the most tokens.
     • Prompt: @report.md <your gpt-5 task>