The five-hour usage limit is returning to Codex and ChatGPT Work for OpenAI's Plus subscribers. The decision, announced by Thibault Sottiaux, the engineering lead for Codex and ChatGPT, marks the end of a several-week experimental period during which users were subject only to a weekly cap. The change takes effect on August 25 and applies to intensive workflows such as automated software development and complex business task management.
Smoothing Compute Load
According to Sottiaux, reinstating this limit is necessary for two main reasons. First, it allows OpenAI to smooth the load on its compute infrastructure while keeping the overall weekly quota generous. Second, it prevents Plus plan users—who are often new or less experienced with intensive AI agent usage—from accidentally burning through their entire weekly allowance in a single session, thereby improving long-term usability and predictability.
Pro and Enterprise Exemptions
The move does not affect all users: Sottiaux clarified that for the coming months, the five-hour limit will not apply to Pro subscriptions costing $100 and $200 per month. Enterprise and Edu plans, which use a separate and more flexible credit-based system, also remain exempt from this restriction. This differentiation reflects OpenAI's strategy in balancing accessibility for new users with the needs of professionals who rely on uninterrupted work sessions.
The Context of Codex's Evolution
This decision fits into a phase of rapid maturation for the OpenAI ecosystem. As reported by AlexTech, the integration of Codex and the launch of ChatGPT Work transformed the platform into an agentic productivity hub. Furthermore, the new reset system for Codex had already introduced 'banked resets', allowing developers to manually manage their quotas. With the return of the five-hour limit, OpenAI is now seeking to stabilize the user experience ahead of autumn, when Codex is set to become a full operating system for business productivity, shifting processing to the cloud.

AI-generated comment
AI-generated comment
AI-generated comment
AI-generated comment