OpenAI Brings Back the 5-Hour Limit for ChatGPT Plus — Pro Subscribers Keep Unlimited Sessions
Starting August 25, Plus users face the restored five-hour usage cap on Codex and ChatGPT Work, six weeks after OpenAI lifted it. The company says it needs to smoothen compute load — Pro tiers at $100 and $200 stay uncapped for months.
OpenAI is restoring the five-hour usage limit for ChatGPT Plus subscribers starting today, August 25 — six weeks after removing it on July 12. The cap applies to the compute-heavy Codex and ChatGPT Work experiences; regular ChatGPT conversations are unaffected. Pro subscribers on the $100 and $200 tiers keep uncapped sessions “for the upcoming months.”
The official rationale comes in two parts. First, compute: the rolling five-hour window lets OpenAI “smoothen the load” on its infrastructure, which in turn lets it keep the weekly allowance generous. Second, user protection: OpenAI says casual Plus users were unknowingly burning through their entire week’s usage in marathon sessions, then hitting a wall with days left on the clock and no idea why.
Both reasons are probably true. Neither is the whole story.
The economics of unlimited agents
The July removal coincided with GPT-5.6 Sol’s efficiency improvements — the model reportedly serves the same workloads roughly 18% cheaper. What OpenAI evidently discovered is that efficiency gains do not survive contact with agentic workloads. A Codex session is not a chat: it spawns long-running autonomous tasks that chew through tokens whether or not the human is even at the keyboard. Give power users unlimited five-hour windows and they will queue agents around the clock, because that is the rational thing to do.
This is the same lesson every AI coding vendor has learned in sequence: Anthropic added weekly caps to Claude Code after users ran it 24/7, Cursor repriced its “unlimited” plan, and now OpenAI walks back its own experiment. Flat-rate pricing and autonomous agents are fundamentally incompatible. The agent does not get tired. The subscription price assumes you do.
What Plus users actually lose
The practical impact depends entirely on usage pattern. The five-hour window is rolling, not a daily allotment — heavy Codex users will now hit a pause, wait for the window to reset, and resume. Weekly limits remain the binding constraint for most people, and OpenAI kept those unchanged. If you never noticed the cap was gone, you will not notice it is back.
For professional users the message is less subtle: the $20 tier is not where OpenAI wants agentic workloads to live. The carve-out for Pro subscriptions makes the segmentation explicit — sustained agent usage is a $100-plus product now. Expect the same stratification everywhere: entry tiers for conversational use, premium tiers priced against the actual compute of running agents in loops.
The frustrating part is the whiplash. OpenAI spent July celebrating milestones with early weekly-limit resets and an uncapped experience, and spent late August quietly re-installing the meter. Free capacity is a promotion, not a policy. Plan your workflows — and your subscription tier — accordingly.