Codex
Codex or Work usage running out fast? Day 12 stretches it by reusing subagent context and caching longer
Tibo's Day 12 ship: OpenAI changed how Codex and ChatGPT Work handle context so the same plan allowance goes further. Subagents reuse context, context stays cached longer, and routing lands more requests on the cache. No setting to change, and no number was given.
What changed
On Day 12 of his daily ship series, Codex lead Tibo (@thsottiaux) said OpenAI made changes to Codex and ChatGPT Work that increase the effective usage you get from your plan. He named three changes. Subagents now reuse context instead of each one starting from scratch. Context stays cached longer. And request routing was improved so more requests hit the cache.
The idea is that cached context is cheaper to process than fresh context, so the same work should take a smaller bite out of your allowance. Your limits didn't change; the work you do should just count for less of them.
Who is affected
People who use Codex or ChatGPT Work on a ChatGPT plan and run into usage limits. It should matter most if you split tasks across subagents, keep long threads going, or come back to the same thread after a pause, since those are the cases where context gets reused or cached.
What to do now
- You don't need to change any setting. The improvement is on OpenAI's side.
- If you split work across subagents, keep doing it. Context they share with the parent thread is now reused rather than sent fresh each time.
- Keep related work in one thread instead of starting a new thread for each small follow-up, so more of your context can come from the cache.
- Watch your usage meter over the next few days on the same kind of work you did last week. That's the only honest way to see how much this helps you.
- If usage still drops faster than you expect, reply to Tibo's post or send Codex feedback with what you were doing. Plenty of replies to his post say usage still feels volatile, so the team is asking for those reports.
What is not confirmed
Tibo didn't give a number, so we don't know how much further your allowance will go, and the gain will depend on how you work. As of Oct 16, the change isn't in OpenAI's changelog or ChatGPT release notes, and DotsFeed hasn't measured it. Which plans and models are covered wasn't stated.
Sources
- Tibo (@thsottiaux) on X, Day 12: https://x.com/thsottiaux/status/2111244596131111253
Sources
Get updates like this every morning
- ① Email
- ② Card on Stripe
- ③ 7 days free
Then $2/month · cancel anytime in one click