BLUF: ChatGPT on the Mac has become dramatically more capable—and dramatically harder to understand at a glance. I hit my weekly usage limit for the first time last week, which pushed me to try Proton Lumo as a backup. Then OpenAI restored the 5-hour limit on August 25th. Here’s what the documentation actually says about how these limits work, and why users shouldn’t need a PhD in product architecture just to read a usage meter.
What Happened Last Week
Let me set the scene. Last week, I hit my weekly ChatGPT usage limit for the first time in months. I’m a heavy user — I run WordPress automation workflows, content pipelines, coding projects, the whole deal. But I’d never actually bumped against that ceiling before.
That moment is what pushed me to sign up for Proton Lumo Plus. Not because I’m abandoning ChatGPT — I still use it daily and it’s genuinely valuable — but because I needed a backup when the meter runs dry.
Then, on August 25th, OpenAI announced they were restoring the 5-hour usage limit for Codex and ChatGPT Work on Plus accounts. This had been temporarily lifted earlier in the summer, and now it’s back. So suddenly I’m looking at my usage panel and seeing:
– 5-hour usage limit — 100% left
– Weekly usage limit — 100% left
– Usage limit resets — 1 available
– Full reset (Weekly + 5 hr)
– An expiration date for that reset
– Buttons for Upgrade plan and Add credits
My reaction was essentially: What exactly am I looking at?
The Mac App Has Been Changing Nonstop
The ChatGPT macOS desktop app has been updating constantly. Here’s what I pieced together from OpenAI’s own changelog and release notes:
On July 9, 2026, OpenAI released a major redesign that merged Chat, Work, and Codex into a single “super app” on macOS. The previous standalone ChatGPT app was renamed “ChatGPT Classic,” and the Codex application became the foundation for the new unified experience.
Since then, August alone brought:
– August 7: Improved file handling, in-app search, and continuity between sessions. GPT-Live voice conversations gained the ability to accept file uploads and reference Project context.
– August 13: A new “Computer History” opt-in feature that logs cross-app and web activity into a timeline. Record & Replay expanded to EU, UK, and Switzerland.
– August 14: Interactive quizzes, enhanced project-memory handling, and seamless switching between typing, dictation, and desktop-focused modes.
– August 20: An Apple Messages plugin (Apple silicon only, opt-in) that can read iMessage/SMS/RCS history and send messages on the user’s behalf.
– August 25: Webhook-triggered scheduled tasks in Work, shared-task capabilities, more flexible limits for free users, and an upgraded Work browser that can complete actions on signed-in sites.
That’s five feature drops in one month. Each one adds capability. Each one shifts the ground underneath you while you’re trying to get actual work done.
What the Usage Meters Actually Mean
I dug through OpenAI’s documentation, community forums, and changelogs to decode this. Here’s what I found:
5-Hour Usage Limit
This is a rolling 5-hour window for active prompting within Codex and ChatGPT Work. It’s not five hours of clock time — it’s a window during which you can send a capped number of messages.
For Plus users, this translates to roughly 15-40 minutes of actual reasoning time depending on model and task complexity. For Pro 5X subscribers ($100/month), it scales up to roughly 1-2.5 hours per window. Pro 20X ($200/month) gets unlimited voice access but not unlimited Codex.
After hitting the limit, the window rolls forward. You wait, buy credits, or upgrade.
Weekly Usage Limit
This is a separate aggregate cap on top of the 5-hour window. You can burn through your 5-hour allocation and still have 80%+ of your weekly limit intact — which is exactly what my usage screen showed. Or you can grind through both if you’re running heavy agentic workflows all week.
Work and Codex Share One Pool
According to OpenAI’s own pricing documentation:
“ChatGPT Work and Codex share usage. ChatGPT Work usage inside ChatGPT uses the same pricing, credits, and usage limits as Codex.”
Translation: There’s one bucket for both products. You don’t get separate allocations.
Voice Is Different (But Not Clearly Different)
ChatGPT Voice on desktop uses a separate, plan-dependent allowance measured in its own rolling 5-hour windows. But — and here’s where it gets confusing — tasks started through Voice use your existing Codex/Work usage budget.
So Voice conversations themselves have separate limits, but Voice-triggered agent tasks eat into your Codex allowance. That distinction is not explained anywhere in the usage panel.
Credits
When you exhaust your included allowances, you start burning through purchased credits at token-based rates. The relationship between included allowance exhaustion and credit consumption isn’t shown visually — you just see an “Add credits” button.
Banked Resets
OpenAI occasionally grants “banked reset” credits through promotions, referrals, or as compensation for outages. According to OpenAI’s Codex documentation:
“Using a full banked reset resets both your 5-hour and weekly usage windows. As a result, your weekly reset date will move to roughly seven days after you redeem it.”
These expire in 30 days. One community tracker counted 36 reset events since the system launched.
File Uploads, Images, and Other Features
“ChatGPT file uploads, image generation, or voice have separate usage limits, reset periods, and banners. For example, if you see banners like ’50 images in the last day’, ‘resets in 720 hours’, reach voice caps, or see upload-limit banners while uploading files in ChatGPT, those limits do not apply to Codex.”
So consuming voice minutes or upload bandwidth does not affect your 5-hour Codex limit. They’re tracked separately. But nowhere in the usage screen does it say that.
The Voice Problem: A Real-World Example
On mobile, I can have an ongoing Voice conversation inside a normal ChatGPT conversation. On my Mac, I’ve encountered situations where activating Voice creates a new Voice/Work/Codex-style conversation rather than simply continuing the existing text conversation in Voice.
This creates an absurd user experience: the technology is sophisticated enough to run agents, manipulate websites, write code, and perform lengthy workflows — yet something as conceptually simple as “continue this same conversation using my voice” can become confusing depending on which part of the application you’re using.
From the documentation, Voice interactions have their own distinct caps and reset schedules that don’t draw from the Codex/Work pool — but Voice-triggered agent tasks do. So the behavior you experience depends on whether Voice is just chatting or actually launching tasks. Good luck figuring that out from the interface alone.
What I’m NOT Saying
Let me be crystal clear: I am not arguing that ChatGPT is bad. Quite the opposite.
ChatGPT has become substantially more useful to me. Codex and agentic workflows are doing real work — WordPress publishing workflows, content production, automation scripts, coding projects. The technology is remarkable. The models are impressive. The automations I’ve built with it pay for the subscription many times over.
My criticism is about product communication and user experience.
Powerful software inevitably becomes more complicated. But complexity underneath the hood doesn’t require confusing terminology on top of the hood.
What the Usage Screen Should Do
If an interface displays “5-hour usage limit,” put a small information button beside it explaining exactly what consumes that allowance.
If Voice behaves differently in Chat versus Work versus Codex, explain that where the Voice button appears.
If a reset changes both weekly and five-hour allowances, explain what happens before the user presses it.
If credits are consumed only after certain included allowances are exhausted, show that relationship visually.
Don’t make customers reverse-engineer their subscription.
The Bigger Picture
This is exactly why THE Tech Voyager exists. Modern technology is extraordinarily capable, but technology companies repeatedly expose internal product distinctions, terminology, and billing structures to ordinary users without translating them into human language.
Users shouldn’t have to understand a company’s organizational chart or product architecture just to understand a button.
The ChatGPT Mac app can now read your iMessages, browse the web with your credentials, index your computer activity, run autonomous agents, and generate code. That’s legitimately incredible. But it also means the stakes for clear communication are higher than ever.
When a product can send messages on your behalf and log everything you do across every app, “what does this button do?” stops being a casual question and starts being a security concern.
Final Verdict
ChatGPT on the Mac has become dramatically more capable — and dramatically harder to understand at a glance. I keep paying for it because it works. The features are real, the models are impressive, and the automations I’ve built with it are genuinely valuable.
But excellence in capability deserves equal investment in clarity. If you’re going to charge customers for sophisticated tools, at least make the bill understandable.
Recommendation: Keep using ChatGPT. It’s worth it. But don’t be surprised when the usage screen confuses the hell out of you — that’s by design, not accident. And when you figure out what the meters mean, document it somewhere. Because the next update will change it again.
Sources and Further Reading
1. Using Codex with your ChatGPT plan – OpenAI Help
2. ChatGPT Pricing Documentation
3. OpenAI restores 5-hour Codex and Work limits – 9to5Mac
4. ChatGPT Release Notes – OpenAI Help
5. ChatGPT & Codex Changelog
6. How to Check Codex Usage and Limits in ChatGPT – J.D. Hodges
7. Understanding the New Codex Limit System – OpenAI Community
8. Free ChatGPT/Codex App Limit Resets – YouTube
9. ChatGPT 5-Hour Limit Removed — July 2026 – explainx.ai
