Codex will spend a saved limit reset, but only when asked

Codex can now spend a banked rate-limit reset by itself, but only after a user asks for one in the conversation and only once the current limit is down to 10% or less, according to the setting’s text shared in a Threads post on October 10. If a user asks while more than 10% is left, or never asks, Codex leaves the reset unused, because the setting’s description says resets are never used automatically.
The toggle is called “Allow Codex to use resets” and sits in the Usage & billing section. Both conditions still have to be met, so the agent only takes over the click. It is unclear whether the 10% threshold is fixed.
OpenAI announced the resets themselves on June 12, when Codex users got the option to save rate-limit resets and use them later. Eligible users on Go, Plus, Pro and Business started with one free banked reset, and the rollout was gradual, so some eligible accounts got the option later than others. A banked reset expires 30 days after it is issued. Once activated, the usage it restores lasts for 7 days.
Before banking existed, the Codex team reset weekly limits for all users to mark major milestones or after big bug fixes, according to a post on OpenAI’s community forum. Anthropic later did something similar in Claude Code: in September it gave Pro, Max and Team users a spare limit reset they could spend any time until October 22.
Up next
Claude hides a heart-in-hand icon behind codename Arborio
Per TestingCatalog, a hidden Claude tab codenamed Arborio carries a heart resting in a hand, and it could bring the sleep and activity analysis from Claude’s mobile apps to the web. Would you hand Claude your sleep and activity data, or keep it in a separate app?
Related stories
- A $200 Codex weekly limit burned in a median 23 hours
- Codex CLI 0.157.0 leaves a daemon running after you quit
- Claude hands paid users a spare limit reset until Oct 22
- ChatGPT and Codex will share one usage pool in December
- OpenAI doubles Pro and Business Premium usage limits
- GPT-6.1 Sol gets 8x speed at $60 per million output tokens
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

