OpenAI removed the rolling five-hour usage restriction that governs ChatGPT and Codex activity for Plus, Pro, and Business subscribers, and reset every account’s usage counter. The change, confirmed by OpenAI product lead Tibo in a post on X, followed what he described as a surge in demand for GPT-5.6 Sol, the company’s most capable model, over the prior two days. The cap removal is temporary. Nothing about its duration has been specified.
Codex, OpenAI’s coding agent, and ChatGPT draw from a shared usage pool that counts both local chat messages and cloud-based agentic tasks. Under the standard system, that pool refills on a rolling five-hour window, with separate weekly ceilings layered on top depending on plan tier and model. Removing the five-hour restriction means Plus, Pro, and Business users can keep working past the point where they would normally be locked out until the window reset.
Tibo also said OpenAI is shipping changes that make GPT-5.6 Sol “more efficient across the board,” cutting how much usage each interaction consumes so the same allotment covers more work. BleepingComputer, which first reported the change, noted that OpenAI has not disclosed the mechanism behind the efficiency gain, though it likely traces to lower token consumption per request.
A temporary cap lift paired with a usage reset is a demand-shaping tool as much as a customer-goodwill gesture. Lifting the five-hour ceiling absorbs a demand spike without OpenAI having to explain why paying customers are getting throttled at a moment when the company wants Codex and GPT-5.6 Sol to look responsive. The reset erases whatever usage customers had already burned, a change that matters more to power users running long agentic sessions than to casual chat users who rarely approach the ceiling anyway.
The word doing the real work here is “temporary.” OpenAI has not said when the five-hour restriction returns, nor whether it comes back at the same threshold, a higher one informed by the efficiency work, or a lower one if the demand surge exposed capacity strain the company would rather not repeat in public. Weekly limits, which OpenAI has confirmed still apply depending on plan and model, remain the backstop regardless of what happens to the five-hour window.
Anthropic has taken a parallel path this month, extending free Claude Fable 5 access and raising Claude Code limits through July 19, so the timing places both major labs loosening usage constraints at once rather than OpenAI acting alone.
For operators, the practical move is to treat this window as a gift, not a baseline. Run the throughput-heavy evals or long Codex sessions deferred because of the five-hour wall while that wall is down and usage counters read zero. Do not build a capacity plan, a pricing assumption, or a customer commitment around the current limits holding. OpenAI has given itself full discretion to reinstate the restriction, and the absence of a stated end date means it could happen without warning.
Reported by Mayank Parmar for BleepingComputer.
