Switching Codex reasoning mid-thread no longer breaks your cache — plus faster Dots calls and new Codex CLI presets

#What changed

On Oct 23, Tibo (Thibault Sottiaux, Codex lead at OpenAI) posted his Day 19 note, which is a set of small updates rather than a usage reset. Changing the reasoning level in Codex, or in Work, partway through a conversation no longer invalidates the prompt cache. Calls in Dots now connect much faster. The Codex CLI got reusable prompt presets, inline model and permission settings, enterprise MCP sign-in, a fix for failures when resuming long sessions, and Windows corporate-proxy authentication. He also said CLI usage has roughly doubled since DevDay.

#Who is affected

Anyone who starts a Codex thread on a low reasoning level and bumps it up for a hard step (or the reverse), Dots users who start calls with their Dot, Codex CLI users who retype the same instructions, and enterprise or Windows users behind a corporate proxy.

#What to do now

Stop opening a new thread just to change reasoning: switch the level in place and keep going, since the cached context now carries over. Start cheap on lower reasoning for exploration, then raise it for the tricky change. In the CLI, update to the latest version and save the prompts you repeat (review checklists, test-and-fix loops) as presets. If long resumed sessions used to fail for you, try resuming again after updating.

#What is not confirmed

These details come from Tibo's posts on X. We have not yet seen them in an official changelog, and exact CLI command names for presets were not shared, so check the CLI's help after updating. The 2X usage figure is Tibo's statement, not independently verified.