Voices
OpenAI quietly merged a 'persistent' reasoning-effort value into Codex, then told Wired it has no plans to ship it
PR #40799 in openai/codex, merged August 26, adds `persistent` to the reasoning-effort protocol and the TypeScript SDK types and surfaces model-advertised persistent reasoning in the TUI. The described behavior is that Codex 'continues working until put to sleep,' extending sessions past today's bounded task runs, but the setting stays in local config and the value actually sent to the API is `disabled`. An OpenAI representative told Wired the company is trying the capability out with no near-term release intent, which makes this a read on direction rather than a usable feature.
↳ Follow the thread