OpenAI's Codex usage limits operate on rolling 5-hour and weekly windows that reset automatically based on a user's first request after the prior period ends, with OpenAI frequently issuing ad-hoc global or banked resets to paid subscribers. These interventions, often announced via @thsottiaux, have occurred roughly every 3–8 days recently and tie directly to model launches such as GPT-6 Sol and Luna, efficiency fixes for overcounting background tasks, and milestones like active user growth. The most recent banked reset landed September 29 following DevDay announcements that introduced Ultrafast tiers, Decisions API features, and Pro plan adjustments effective late October. Traders watch these patterns because resets expand effective access without changing base allowances, while competitive model improvements and subscription tweaks can shift demand and quota pressure.