OpenAI will restore a 5-hour usage limit for ChatGPT Work and Codex on the ChatGPT Plus plan starting August 25 (US time). This reverses part of a July 12 change that had temporarily removed the short-term cap across Plus, Business, and Pro plans—but the reinstatement now applies only to Plus. Pro 5x and Pro 20x will keep the 5-hour limit disabled for the next several months. This is not an announcement about increasing or decreasing weekly usage totals overall; rather, it's an operational adjustment designed to smooth out compute demand while preventing Plus users from exhausting a full week's quota in a short burst of activity.
From a Temporary Removal in July to a Plus-Only Restoration
On July 12, shortly after the public release of GPT-5.6 Sol, OpenAI temporarily removed the 5-hour limit from Plus, Business, and Pro plans amid surging demand. OpenAI Support reaffirmed on July 28 that these three plans were the ones affected. Because the weekly cap remained in place, users were less likely to be blocked during short sessions—but running long tasks or multiple concurrent agents could still burn through an entire week's allowance early on.
About six weeks later, the treatment of individual plans is diverging. Thibault Sottiaux, who oversees Codex and ChatGPT at OpenAI, announced that the 5-hour limit will return for Work and Codex on Plus. The $100/month Pro 5x and $200/month Pro 20x plans are excluded, and this exclusion will remain in place for the next several months. Sottiaux did not describe this as a permanent measure, nor did he give an end date. OpenAI has not disclosed regional rollout timing or whether the change will be staged.
The announcement made no mention of changes—or lack thereof—for Business, Enterprise, or Edu plans. As a result, the restoration for Plus should not be read as a uniform change extending to other plans. What is confirmed is this: the short-term cap is coming back for Plus, and it will remain disabled for the two Pro tiers for now.
By limiting the restoration to Plus, OpenAI has also created a split among individual plans in terms of the time window governing usage. Sottiaux explained that Plus tends to include more newer or occasional users. Pro plans, by contrast, are marketed as offering 5x or 20x the usage of Plus. A short-term pause functions as an early warning for overconsumption on Plus, while Pro users retain the ability to keep working without interruption. Beyond the price difference, this also creates a difference in the time window within which the same amount of compute can be consumed.
Smoothing Compute Demand While Curbing Weekly Quota Burnout
Sottiaux cited two reasons for the change. First, the 5-hour limit helps smooth out spikes in compute demand, making it possible to keep weekly usage allowances relatively generous. Second, it reduces the risk that newer or occasional users unknowingly exhaust their weekly quota without realizing it. The company appears to be using the same short-term limit to manage both its own infrastructure load and users' budget awareness.
Under OpenAI's standard system, local messages and cloud chats on a ChatGPT plan draw from the same 5-hour limit, and a weekly cap may also apply on top of that. Moreover, ChatGPT Work and Codex share usage, credits, and limits. If a user researches materials and produces a deliverable in Work, then fixes code in Codex, both actions draw from a single shared allowance rather than two separate budgets tied to different products.
The reinstatement of the 5-hour limit, therefore, is not a substitute for the weekly cap. It reimposes a short-term ceiling on usage and breaks consumption into smaller increments before it flows into the weekly total. Since OpenAI has not disclosed the fixed weekly cap figures or any rate of increase or decrease tied to this change, it cannot be said that "the weekly total is increasing in exchange for the return of the 5-hour limit."
The effect of smoothing compute demand is also distinct from actually increasing available compute resources. By curbing the concentration of heavy processing within short periods, OpenAI can lower peaks in concurrent usage. According to Sottiaux, this in turn allows the company to keep weekly allowances more generous. The 5-hour limit is a constraint from the user's perspective, but for OpenAI it also serves as a lever for calibrating how much compute can be bundled into a flat-rate plan.
"5 Hours" Does Not Mean Continuous Usage Time
The "5-hour limit" is not a contract guaranteeing five continuous hours of usage after logging in. OpenAI presents ranges of local message counts by model, and even within Plus, the guideline figures differ significantly between GPT-5.6 Sol and Luna.
| Plan | Monthly Price | Sol | Terra | Luna | Current Status of 5-Hour Limit |
|---|---|---|---|---|---|
| Plus | $20 | 10–100 | 25–200 | 250–2,000 | Restored |
| Pro 5x | $100 | 50–500 | 125–1,000 | 1,250–10,000 | Remains disabled |
| Pro 20x | $200 | 200–2,000 | 500–4,000 | 5,000–40,000 | Remains disabled |
The figures in the table represent guideline ranges for the number of local messages that can be sent within a 5-hour window—they are not fixed counts. Consumption per message varies depending on task size and complexity, as well as the context the model retains, the amount of reasoning performed, and tool usage. Search and caching states also play a role, meaning that instructions of the same length may not deplete the allowance at the same rate.
The Pro figures are the guideline numbers that OpenAI's official pricing page uses to describe the standard 5-hour limit for comparison purposes. They do not imply that Pro users will be capped at these counts during the current exception period. Meanwhile, for the weekly cap, OpenAI states only that "an additional limit may apply," without publishing fixed figures by plan. Users can compare guideline numbers for the short-term limit by model, but for the weekly limit, they have no choice but to manage their pace based on the remaining-usage display.
Cloud-based chats use GPT-5.6 Sol and may consume more than local messages do. Counting a long cloud-based task and a lighter local edit as equivalent "single" actions does not mean they place equal strain on the usage allowance. It's more accurate to think of the 5-hour limit not as a countdown timer measuring elapsed time, but as a budget that caps the amount of compute used within a given period.
Plus Users Must Now Watch Both Short-Term Wait Times and Weekly Balances
For Plus users, the possibility of wait times tied to the 5-hour limit is returning—unlike during the temporary removal period, when users could focus solely on the weekly cap and batch long tasks together. On the other hand, the short-term limit makes it easier to notice overuse early, reducing the risk of losing the entire weekly allowance in the first half of the week. OpenAI advises users to check their remaining balance via the usage dashboard or the /status command in Codex CLI.
Even after hitting the limit, a turn already in progress can be completed within reasonable-use bounds. If users need to continue working, both Plus and Pro allow the purchase of additional credits, or users can switch to smaller models such as Terra or Luna to reduce consumption. However, having more options after exceeding the short-term limit does not mean additional usage becomes free.
How work is structured also matters. On Plus, running a long cloud-based task and local work simultaneously within the same 5-hour window will draw on the short-term limit at the same time. OpenAI positions Sol for complex reasoning and advanced coding, Terra for everyday practical work and development, and Luna for classification, extraction, and targeted coding tasks. Which model a task is routed to affects not only the weekly remaining balance but also short-term interruption time. Since prompt length alone cannot predict consumption, users need to allocate their work while watching their actual remaining balance.
Pro 5x and Pro 20x are more likely to avoid short-term interruptions, but they leave more room to burn through the weekly allowance at a faster pace. Plus, conversely, has work segmented by the short-term cap but is better protected against depleting its weekly balance. What OpenAI should clarify going forward is how the restored limit actually feels in practice for Plus users, and when the "several months" exclusion period for Pro will end. Once that becomes clear, it will be possible to judge whether the 5-hour limit is merely a stopgap measure for load management, or a step toward a more permanent design that differentiates individual plans by usage pattern.