OpenAI has temporarily removed the rolling five-hour usage window from GPT-5.6 Sol sessions inside Codex and ChatGPT Work. The change, announced by product lead Tibo Sottiaux on July 12, applies to Plus, Pro, and Business subscribers and comes after two days of what Sottiaux described as intense activity following the model’s July 9 launch.

Weekly usage caps remain unchanged. The policy shift doesn’t grant unlimited access — it just untethers the hourly constraint so that paid users can spend more of their weekly allocation in a single, uninterrupted block of work.

The short leash is gone — but there’s a longer one

Codex and ChatGPT Work are purpose-built for complex, often lengthy tasks. Codex tackles software development: editing repositories, running commands, reviewing changes, and orchestrating cloud jobs. ChatGPT Work handles research and finished deliverables. Both draw from a shared “agentic usage” pool on plans that include them.

Until now, that pool was governed by two distinct metering layers: a rolling five-hour window that reset continuously, and a broader weekly cap that resets on a fixed schedule. The five-hour limit became a roadblock almost immediately after GPT-5.6 Sol shipped. A developer refactoring a large codebase or a researcher feeding a substantial document collection into Work could exhaust the hourly allowance long before touching the weekly ceiling.

With the temporary removal, users can concentrate their weekly capacity into marathon sessions. You still have a finite bucket — OpenAI hasn’t published exact weekly ceilings for each plan publicly — but you decide how to distribute that bucket across the week. No more mid-task interruptions because a short-cycle timer ran out.

The removal applies exclusively to Codex and ChatGPT Work. Standard ChatGPT conversations, API calls, and free-tier accounts operate under their own separate limits. API customers continue to be governed by rate limits and usage-based billing, not agentic quota pools.

Alongside the limit removal, OpenAI made several other adjustments:

  • Inference optimizations that Sottiaux said should give users roughly 10 percent more effective usage from their existing weekly allowance. The company didn’t provide technical detail, so whether the gain comes from lower token counts, model compression, or other changes remains speculative.
  • A temporary reduction of the product context setting from 372,000 tokens to 272,000 tokens after discovering the larger value charged more quota than intended. This isn’t a model-level change — GPT-5.6 Sol’s full 1.05-million-token context window is untouched — but it affects how much context Codex and Work expose and account for. OpenAI says the 372,000-token setting will return after additional engineering.
  • Corrections to multi-agent consumption. Launch experiments with reasoning-effort values and multi-agent operation at High and Extra High settings were sometimes overcharging quota. Parallel workers can speed up a task by dividing the work but, as the company acknowledges, they also burn through allowance faster. The corrected accounting should bring those costs closer to the intended level.

Banked resets: a one-time safety valve

OpenAI also distributed a banked reset to every affected account — initially targeting about 500,000 users, then expanding to a claimed seven million after fixing a redemption bug and enabling web and mobile access. A banked reset can be stored and redeemed at any time to refill the weekly allowance. Once used, the account returns to its regular allowance and renewal schedule.

These resets are not permanent quota increases. They are a one-shot credit, and in some cases they may be tied to referral promotions. Users should check the terms displayed in their own usage summary rather than assume a universal offer. Plus and Pro subscribers who exhaust their included agentic usage can also purchase additional credits; the same credit balance can be used for Codex, ChatGPT Work, and ChatGPT for Excel, with optional automatic top-ups.

For developers, researchers, and admins: what the shift means

Developers. The immediate win is session continuity. A repository migration, multi-file refactoring, or security audit that spans six or seven hours no longer requires planning around a five-hour reset window. You still need to monitor the weekly meter, especially if you’re working with large contexts or running parallel agents. The removal also means you’re less likely to lose unsaved progress when the short-cycle clock expires mid-task. When the five-hour window eventually returns — and OpenAI hasn’t said when — you’ll want to watch for notification of its reappearance and plan accordingly.

Researchers and knowledge workers. ChatGPT Work sessions that analyze dozens of documents or generate complex reports can now run longer without interruption. The 10 percent efficiency gain, if it materializes evenly, could mean an extra document or two per week under the same allowance. But the temporary context reduction to 272,000 tokens may pinch if your workflow routinely loads very large documents. You might need to split inputs into smaller chunks until the full context setting is restored.

IT administrators and team leads. Forecasting just got trickier because capacity has become a mix of baseline allowance, one-time resets, and paid credits. A team can gain temporary headroom without any durable increase in its recurring entitlement. Individual task costs vary significantly based on model selection, reasoning level, and whether work runs locally or in the cloud. Keep an eye on account-level usage dashboards and communicate the temporary nature of the relief to your teams. If you’re considering credit purchases, check whether auto-top-ups are enabled and what threshold triggers them. OpenAI’s published documentation (linked below) explains the credit system in more detail.

Everyone else. If you use ChatGPT only for standard chatbot interactions, nothing changes. Your limits remain plan-dependent and are unaffected by the changes to Codex and Work. API developers continue under their own rate limits.

Why a five-hour clock was so disruptive

The short-cycle limit was designed to prevent a single user from consuming their entire weekly allowance in one burst, ensuring availability across the subscriber base. But it collided badly with the nature of agentic workloads. A single prompt in Codex can launch multiple background workers, scan thousands of lines of code, execute shell commands, and iteratively revise output. The compute cost isn’t linear: a five-hour session for one developer might accomplish far more than the same five hours for another, depending on the task.

OpenAI’s support documentation already acknowledged this variability, noting that consumption depends on the model, task complexity, session length, and whether work is local or cloud-based. The quick route to the hourly cap after GPT-5.6 Sol’s release highlighted just how different agentic usage is from simple chat. Developers hit the limit fast and complained loudly, prompting the temporary removal.

This isn’t the first time an AI provider has had to retune limits for more demanding use cases. Google’s Gemini products, for instance, maintain short-cycle limits that refresh every five hours alongside weekly restrictions, according to Google’s own support pages cited by WinBuzzer. Anthropic, meanwhile, recently extended its promotional access to Claude Fable 5 and kept weekly Claude Code allowances 50 percent above normal through July 19. These aren’t direct comparisons — Google still has the short cycle, and Anthropic’s move was a time-limited promotion — but they illustrate how quickly consumption patterns shift when models become more capable.

No evidence from the announcements suggests OpenAI’s decision was a reaction to either competitor. More likely, the company simply underestimated how fast agentic usage would chew through the five-hour window, and the temporary fix gives its team time to recalibrate.

What you should do right now

  1. Check your account for a banked reset. Log into your OpenAI account, navigate to the usage or settings area, and look for a banked reset offer. You can hold it until you really need it. If you don’t see it and believe you should, consult OpenAI’s help portal — some accounts may have experienced the redemption bug that has since been fixed.
  2. Understand your weekly cap. OpenAI hasn’t published a single number for all plans, so you’ll need to monitor your usage dash. Plan setups vary; Pro accounts typically receive a higher agentic allowance than Plus, and Business plans may have pooled quotas. The removal of the five-hour window doesn’t change the total weekly capacity — only when you can spend it.
  3. Choose reasoning levels deliberately. High and Extra High multi-agent settings consume allowance faster because they spin up parallel workers. If a task doesn’t require that depth, consider dropping down a tier to stretch your weekly pool further. The recent corrections to multi-agent billing should help, but heavy parallelism still costs more.
  4. Watch the context window. The product-level context setting is currently 272,000 tokens, not 372,000. If you’re working with exceptionally large codebases or document sets, you may hit this ceiling before the task is complete. Splitting work into smaller chunks can avoid errors. This is expected to be lifted once OpenAI finishes its fixes, so check release notes regularly.
  5. Monitor official channels for the limit’s return. OpenAI hasn’t said when the five-hour window will be reinstated. Bookmark the official Codex and ChatGPT Work documentation, or follow Sottiaux’s updates if you use those platforms. The relief is temporary, and you’ll want to plan your large sessions accordingly when the clock comes back.
  6. Consider buying credits proactively if you run close to weekly caps frequently. Credits let you continue working beyond the included allowance. They’re pooled across Codex, ChatGPT Work, and ChatGPT for Excel, and can auto-replenish. Read OpenAI’s credit documentation to understand pricing and triggers before you rely on them under deadline pressure.

What comes next

The removal is a stopgap, not a permanent redesign. OpenAI’s next moves will likely depend on whether the efficiency optimizations and cost-accounting corrections are enough to keep users from hitting the weekly cap prematurely. If the 10 percent gain holds, some users may not notice the difference, while those on the edge might still need to manage their sessions carefully.

The restoration of the 372,000-token product context setting will be an early indicator. When it lands without the earlier billing bug, users will get back the full working context they expected. The return of the five-hour window is the bigger question — if the corrected accounting smooths out usage curves, OpenAI might adjust the window rather than bring it back unchanged.

For now, paid users have a rare opportunity to structure their most demanding GPT-5.6 Sol work around the task itself, not around an arbitrary clock.