Does switching Fast Speed on and off during a session invalidate the prompt cache in Codex app?

[wilkice]

Does switching Fast Speed on and off during a session invalidate the prompt cache in Codex app?

I have been using [mattpocock/skills](https://github.com/mattpocock/skills) frequently recently.

My usual workflow in the Codex app is:

1. Enable Fast Speed at the start of a session for discussion and planning.
2. Switch to Standard Speed when I need Codex to implement or rewrite code.
3. Sometimes switch back to Fast Speed after code generation when I need more discussion.
4. Later switch back again to Standard Speed for further code changes.

So within a single session, I may switch between Fast Speed and Standard Speed multiple times.

My question is:

When using the Codex app, does switching Fast Speed on or off invalidate the prompt cache?

Does this behavior apply to all GPT models, or does it vary by model?

Environment

Codex app: latest macOS version
Subscription: ChatGPT Plus
Model: GPT-5.5

What I found

In Claude Code’s documentation, it says that since Claude Code v2.1.86, if Fast Mode is enabled first, then turning it off or turning it back on later still preserves the cache.

I would like to know whether Codex app has similar behavior for GPT models.

I checked the OpenAI prompt caching documentation, but I could not find information specifically about how Speed mode switching in Codex app affects prompt cache behavior:

https://developers.openai.com/api/docs/guides/prompt-caching

Reference from Claude Code:

https://code.claude.com/docs/en/prompt-caching#turning-on-fast-mode

[l1ve709]

Not necessarily, it depends on where the Speed setting lives in the request, not just whether you toggle it.

Prompt caching (on both OpenAI and Anthropic's APIs) works on an exact-prefix match...