ChatGPT Codex Just Lost Its 5-Hour Leash — Here's What Actually Changed
OpenAI has temporarily pulled the rolling 5-hour usage window on Codex and ChatGPT Work for Plus, Pro, and Business plans. The weekly cap is still there. Here's what's confirmed, what isn't, and how to work around it.
If you've hit OpenAI's "You've reached your usage limit" wall mid-session, you'll want to read this. On July 12, 2026, OpenAI's Codex product lead announced a change to how Codex and ChatGPT Work meter usage — and it's the biggest shift to the rate-limit system since the products launched.
Below is a plain breakdown of what changed, what stayed the same, and what OpenAI has and hasn't confirmed. We're sticking strictly to what's been officially stated or reported by credible outlets — anything unconfirmed is flagged as such.
OpenAI has temporarily removed the rolling 5-hour usage window for Codex and ChatGPT Work on Plus, Business, and Pro plans. Your weekly usage limit — which was already running in parallel — is now effectively the only ceiling on how much you can work in a session. OpenAI has explicitly called this change temporary and has not said when, or whether, the 5-hour window will return.
What is ChatGPT Codex?
Codex is OpenAI's coding agent, available inside ChatGPT, as a CLI, and as a desktop app for macOS and Windows. It handles local coding sessions and cloud-based background tasks — writing code, reviewing pull requests, and running multi-step agent workflows — and it's bundled into ChatGPT's Free, Go, Plus, Pro, Business, and Enterprise plans, each with different usage ceilings.
Historically, Codex and ChatGPT Work usage has been governed by two separate meters running at the same time: a short rolling window and a longer weekly cap. That dual-meter system is exactly what's now in flux.
The previous 5-hour usage window
Under the standard system — which is still what OpenAI's own Codex pricing page documents as the baseline — local messages and cloud tasks shared a rolling 5-hour usage allowance. Hit that ceiling, and Codex would stop you cold until the window rolled over, regardless of how much of your separate weekly allowance you had left.
That mismatch was the most common complaint: a long agentic session, or a heavier model like the newer "Sol" variant of GPT-5.6, could exhaust the 5-hour bucket in a single sitting while plenty of weekly quota remained untouched. Developer Theo, a vocal Codex user, described burning through significant token spend on a $200/month Codex Pro subscription and still finding it easy to hit the short-window limit.
What changed: the weekly-only usage limit
On July 12, 2026, OpenAI Codex product lead Tibo Sottiaux posted three updates on X, following two days of heavy demand after the July 9 general release of the GPT-5.6 model family:
Three important updates: temporarily removing the 5-hour usage limit restriction for all Plus, Business, and Pro plans; rolling out efficiency changes to GPT-5.6 Sol; and landing a usage reset after crossing 6 million active users. — Paraphrased from Tibo Sottiaux (@thsottiaux), OpenAI Codex, July 12, 2026
In practice, that means Codex and ChatGPT Work users on Plus, Business, and Pro no longer see the short-cycle counter blocking them mid-session — only the weekly limit applies for now. OpenAI paired the change with a few related adjustments:
Efficiency pass
OpenAI says inference optimizations should make GPT-5.6 Sol consume roughly 10% less usage for the same work, though the company hasn't published exact figures.
Context window rollback
A product context setting was temporarily reduced from 372,000 to 272,000 tokens after OpenAI found the larger setting was charging more quota than intended. OpenAI says it plans to restore the higher figure later.
Banked resets
OpenAI rolled a savable usage reset out to roughly 500,000 users initially, then extended it to web and mobile and confirmed that every account among a stated 7 million active users had received at least one reset.
Multi-agent tuning
OpenAI said it was correcting excess usage draw from parallel multi-agent workers running at high and extra-high reasoning effort, which had been consuming more compute than intended.
OpenAI has described this as temporary at every step and has not committed to making it permanent. The official Codex pricing page still lists the 5-hour window as part of the standard usage-limit structure. Treat the current state as an active window of relief layered on top of the normal system — not a confirmed permanent redesign.
How the weekly reset actually works
The weekly limit is a separate, longer-running allowance that was already stacked on top of the 5-hour window before this change — it hasn't been removed, and it's now the main thing to watch. A few practical points worth knowing:
- The weekly cap doesn't scale in a straight line with the message counts Codex shows you — heavier agent loops with large context reads can burn a disproportionate share of the weekly budget even if a short-window gauge looks healthy.
- You can check remaining weekly capacity in the Codex usage dashboard, or by running
/statusinside an active Codex CLI session. - Resets have not been happening on a fixed calendar cadence during this launch period — OpenAI issued manual top-ups multiple times in the days around July 12–14, in addition to the standard rolling weekly cycle.
- Once your weekly allowance is exhausted, Plus and Pro users can buy additional credits to keep working rather than waiting out a reset.
Timeline of the change
-
JUL 9, 2026
OpenAI generally releases the GPT-5.6 model family, including the "Sol" variant used in Codex and ChatGPT Work.
-
JUL 10–11, 2026
Demand roughly doubles in 48 hours. Users report hitting the 5-hour window repeatedly despite unused weekly quota; OpenAI issues an apology-style acknowledgment of the strain.
-
JUL 12, 2026
Tibo Sottiaux announces the temporary removal of the 5-hour limit for Plus, Business, and Pro, plus efficiency changes and a usage reset tied to crossing 6 million active users.
-
JUL 13, 2026
OpenAI denies that reasoning-budget changes were a covert nerf, promises roughly 10% more usage headroom from inference savings, and extends banked resets toward 500,000 users.
-
JUL 14, 2026
OpenAI confirms every account among a reported 7 million active Codex and ChatGPT Work users has received at least one banked reset.
-
Ongoing
No date has been given for when, or whether, the 5-hour window returns. OpenAI continues to describe the change as temporary.
Benefits of the current setup
- Longer uninterrupted sessions. Multi-hour agentic runs, large refactors, and long debugging sessions no longer get cut off by a short-cycle counter mid-task.
- Better fit for bursty work. If your coding happens in occasional long stretches rather than steady daily use, you're no longer penalized for concentrating it.
- More effective usage per session, thanks to the concurrent efficiency improvements to GPT-5.6 Sol, on top of the limit change itself.
- Extra breathing room from banked resets, which stack with the removed short-window cap during this period.
Drawbacks and open questions
- It's explicitly temporary. Nothing here is a confirmed permanent change, and OpenAI has given no timeline for reintroducing the 5-hour window.
- The weekly cap is still finite — and easy to misjudge. Because it's tracked differently than the old short-window counter, a healthy-looking 5-hour gauge historically didn't guarantee healthy weekly capacity, and heavy multi-agent sessions can drain the week faster than expected.
- Resets during this period have been irregular, issued manually around specific announcements rather than on a predictable schedule.
- The context-window rollback (372K → 272K tokens for the relevant product setting) is a real, if temporary, reduction some users will notice on large-context tasks.
Before vs. now, side by side
| Aspect | Standard system (documented) | Current state (temporary) |
|---|---|---|
| 5-hour rolling window | Active | Removed for now |
| Weekly usage limit | Active | Still active — main constraint |
| Product context setting | 372,000 tokens | 272,000 tokens (temporary rollback) |
| Usage reset cadence | Rolling weekly cycle | Rolling cycle plus manual top-ups |
| Official status | Standard, documented policy | Announced as temporary, undated |
Tips for working within the current limits
- 01Check
/statusin the Codex CLI or the usage dashboard regularly — the weekly number is now the one that matters most. - 02Switch routine, low-complexity tasks to GPT-5.4-mini; OpenAI estimates this can extend local-message usage by roughly 2.5–3.3x compared with a flagship model.
- 03Trim your
AGENTS.mdfiles and disable MCP servers you aren't actively using — both add context that eats into your quota. - 04Front-load demanding agent work early in the week while a reset is fresh, rather than assuming the short-window relief means unlimited daily capacity.
- 05Don't assume this state is permanent — plan critical deadlines as if the 5-hour window could return without much notice.
Key takeaways
- OpenAI temporarily removed the rolling 5-hour Codex/ChatGPT Work limit for Plus, Business, and Pro plans on July 12, 2026.
- The weekly usage limit was never removed — it's now effectively the only cap in place.
- The change came bundled with efficiency improvements, banked resets, and a temporary context-window rollback.
- OpenAI has not said when, or if, the 5-hour window will return.
- Free, Go, and Enterprise/Edu plans follow different usage structures not covered by this specific change.
Frequently asked questions
Is the 5-hour usage limit gone for good?
Not confirmed. OpenAI has repeatedly described this as a temporary change and hasn't given a date for restoring the 5-hour window or a commitment to keep it removed.
Does this affect the ChatGPT Free or Go plans?
OpenAI's announcement specifically named Plus, Business, and Pro plans for Codex and ChatGPT Work. Free and Go tier limits weren't addressed in the same statement.
Does this apply to API usage?
No. The change is specific to Codex and ChatGPT Work usage inside ChatGPT plans. Usage through an API key continues to be billed at standard API rates, separate from these plan-based limits.
Why does my weekly limit still feel like it drains fast?
The weekly cap is tracked differently than the old short-window message counter — heavier agent sessions with large context reads can consume a large share of weekly quota even when a short-window gauge would have looked fine. Watch your weekly percentage directly rather than assuming a healthy-feeling session means healthy weekly capacity.
What happens if I hit the weekly limit now?
Plus and Pro users can purchase additional credits to keep working without upgrading plans, or switch to GPT-5.4-mini to extend remaining usage. Business, Edu, and Enterprise customers on flexible pricing can purchase additional workspace credits.
How does this compare to Gemini and Claude's usage limits?
As of this writing, Google's Gemini Apps retain both five-hour and weekly limits according to Google's own support documentation. Anthropic has run separate, fixed-term promotional extensions to paid-plan limits rather than removing a limit tier outright. These aren't directly equivalent policies, and none of the companies has stated their change was a response to a competitor.
Conclusion
The headline is straightforward: for now, Codex and ChatGPT Work users on Plus, Business, and Pro plans are working against a single weekly ceiling instead of two overlapping ones. That's a genuine, welcome change for anyone who's had a long session cut short by a short-cycle counter. But "temporary" is doing real work in OpenAI's own language here — there's no confirmed end date, and no promise the 5-hour window won't come back. Until OpenAI says otherwise, the safest approach is to enjoy the extra flexibility while keeping a closer eye on your weekly number than you might have needed to before.
References
- OpenAI Codex — Pricing and usage limits documentationchatgpt.com/codex/pricing
- OpenAI Help Centerhelp.openai.com
- Tibo Sottiaux (@thsottiaux) — July 12, 2026 announcement thread on Xx.com/thsottiaux
- BleepingComputer — "OpenAI temporarily relaxes GPT-5.6 Sol usage limits"bleepingcomputer.com
- Digital Trends — "OpenAI just took the handcuffs off your ChatGPT Work and Codex usage limits, at least for now"digitaltrends.com
- WinBuzzer — "OpenAI Eases GPT-5.6 Usage Limits, Keeps Weekly Caps"winbuzzer.com
- OpenAI Developer Community forum thread on the 5-hour reset removalcommunity.openai.com
