On July 12, 2026, OpenAI temporarily removed the 5-hour usage-limit restriction for ChatGPT Plus, Pro, and Business users working with Codex and ChatGPT Work.

The change followed a sharp demand spike around GPT-5.6 Sol, Codex, and ChatGPT Work. OpenAI product lead Tibo Sottiaux said on X that the previous 48 hours for Codex and ChatGPT Work had been intense, then outlined three changes: temporarily removing the 5-hour restriction, making GPT-5.6 Sol more efficient, and landing a usage reset.

This was not an unlimited-use announcement. The short rolling usage window was relaxed for eligible paid plans, while weekly usage limits could still apply.

Update: the five-hour window is documented again

Updated July 30, 2026: OpenAI's current Codex pricing documentation again lists local messages and cloud tasks as sharing a five-hour window, with additional weekly limits possible. That footnote appears under the published plan tables for Plus, Pro, and Business.

Two things are worth separating. The July 12 announcement did not carry an end date, and OpenAI's pricing page does not state when the exception ended. So the honest answer to "is the 5-hour limit gone?" is: OpenAI's published limits currently describe a shared five-hour window, and the temporary relief announced in mid-July is no longer reflected in that documentation.

For an active account, the pricing page and the in-product Codex usage summary are stronger evidence than the July announcement. A user who reaches a limit may be able to add credits, wait for the allowance to reset, apply an available banked Codex rate-limit reset, or upgrade, depending on the plan.

What changed in July

Before the temporary change, OpenAI's Codex pricing page described local messages and cloud tasks as sharing a five-hour window, with additional weekly limits possible. That meant a user could hit the short-window gate even if they had not fully exhausted their longer weekly allowance.

The temporary update removed that shorter interruption for Plus, Pro, and Business users, according to Sottiaux's post and multiple reports covering the announcement.

For heavy users during the exception, the effect was simple: Codex and ChatGPT Work sessions were less likely to stop because the rolling 5-hour bucket had run out. Users could still run into weekly limits, plan-specific allowances, credit limits, or other fair-use controls.

Why OpenAI did it then

The timing mattered. OpenAI had recently expanded GPT-5.6 Sol across ChatGPT Work and Codex, and the model was aimed squarely at higher-effort agent work: coding, research, computer use, and long-running workflows.

Those are exactly the tasks that burn usage quickly. A single ambitious Codex task or ChatGPT Work run can involve large context windows, tool calls, file reads, browser work, generated assets, and long reasoning traces. Users do not experience that as a neat number of messages. They experience it as "I asked the agent to keep working, and the limit arrived before the job was done."

The temporary removal was OpenAI's short-term answer to that friction.

Sottiaux also said OpenAI was rolling out changes to make GPT-5.6 Sol more efficient, so the model would consume less usage for the same kind of work. The announcement did not quantify the exact impact.

What applies now

The important caveat is that the July 12 change was explicitly temporary, and the five-hour gate was never the only limit.

OpenAI's pricing page says usage depends on model choice, task complexity, local versus cloud execution, context size, reasoning, tool use, retrieval, and caching. Tasks that look similar can consume different amounts of allowance.

The same page describes plan-based allowances, says additional weekly limits may apply, and notes that users approaching a limit can switch to a smaller model to stretch the remaining allowance.

So the safer reading is:

  • OpenAI temporarily removed the short 5-hour restriction on July 12;
  • the currently published limits again describe a shared five-hour window;
  • weekly limits remain relevant;
  • OpenAI said GPT-5.6 Sol was being made more efficient;
  • neither the announcement nor the pricing page identifies the exception's precise end time.

Why this matters for agent work

For ordinary chat, a 5-hour window is inconvenient. For agentic work, it can break the rhythm of the job.

Codex and ChatGPT Work are built around longer tasks: reviewing code, making changes across a repository, writing reports, creating deliverables, using desktop apps, and continuing in the background. Those workflows do not map cleanly to a fixed short-window counter.

That is why this episode matters more than a simple quota tweak. It suggests OpenAI is still tuning the subscription model around how people actually use agents. The more OpenAI pushes ChatGPT into real work, the more important it becomes for limits to match project flow rather than chat-message habits.

It also came during a competitive moment. Anthropic has been extending access to Claude Fable 5 for paid users while it manages demand and capacity. OpenAI's move looked like the same kind of subscriber-relief strategy: keep users working during a high-demand launch week instead of making them stop right when they are testing the new model hardest.

Our take

This was good news for heavy Codex and ChatGPT Work users, but it was a temporary relief valve, not a new permanent entitlement.

The most useful immediate effect was fewer interruptions during long GPT-5.6 Sol sessions. The bigger long-term question is whether OpenAI replaces the five-hour window with a clearer usage model that feels less surprising to people doing real work.

For now, treat OpenAI's current pricing table and your in-product usage summary as the operational source of truth. Five-hour and weekly allowances can both matter, and high-context agent work can consume usage quickly.