Agents AI

Update
agents

OpenAI Reinstates 5-Hour Usage Cap on Codex and ChatGPT Work for Plus Users

OpenAI brought back a five-hour rolling usage window for Codex and ChatGPT Work on the $20/month Plus plan starting August 25, six weeks after lifting it, citing server load and accidental quota burn by casual users.

AgentsAI NewsroomAugust 27, 20262 min read

OpenAI has reinstated a five-hour rolling usage limit on Codex and ChatGPT Work for subscribers on its $20-a-month ChatGPT Plus plan, effective August 25, according to multiple reports citing an announcement from Thibault Sottiaux, OpenAI's technical lead for Codex and ChatGPT Work, posted on X on August 24. The cap had been lifted for roughly six weeks after OpenAI removed it in mid-July.

What changed

The five-hour window limits how much of a Plus user's weekly Codex or ChatGPT Work quota can be consumed in any single stretch, rather than reducing the total amount of usage available per week. In practice, that means a user can no longer burn through an entire week's allotment in one long coding or research session — usage now has to be spread out across rolling five-hour blocks. OpenAI said the change applies only to the $20/month Plus tier; the $100 and $200 monthly Pro plans will not have the five-hour cap reactivated, according to the reports.

Why OpenAI brought it back

Two reasons were given for the reversal. First, spacing usage into five-hour windows helps OpenAI smooth demand on its serving infrastructure, avoiding the large simultaneous spikes that uncapped access can create. Second, OpenAI said the cap protects newer or more casual users: during the uncapped period, some Plus subscribers unintentionally exhausted their entire week's quota during one extended session and were then locked out of Codex and ChatGPT Work for the rest of the week without realizing what had happened.

Why it matters

The reversal is a reminder that coding agents like Codex remain expensive to run at scale, and that usage-based tension between generous access and infrastructure cost hasn't gone away even as agent products mature. It also illustrates a pattern across the industry this year: labs experimenting with looser limits to compete on usability, then dialing them back once real-world usage patterns reveal the cost or reliability trade-offs — a dynamic Anthropic has cited for its own Claude Code capacity, and one likely to keep recurring as coding-agent adoption grows.

AI-assisted reporting, overseen by the AgentsAI team. Spotted an error? Let us know.