ChatGPT Pro 20x ($200) gives you four times the usage of Pro 5x ($100), not twenty times, because both numbers are measured against Plus. That also means a Full reset on 20x refills a four-times-bigger tank. But during the reset-heavy Astra launch in September 2026, the real bottleneck turned out to be something else: having enough free time to keep the AI busy.
You want to upgrade and run more AI work. The payment or plan switch gets stuck. Meanwhile, usage resets are being handed out all around you.
At some point the joke writes itself:
“Just lend me Tibo’s computer. Unlimited, please.”
To be clear, this is not about borrowing someone else’s account or internal access. It is about what happens when you push Work and Codex hard during the GPT-6 Astra launch: your usage allowance starts to feel less like a message quota and more like research capacity.
And once you dig in, the question is bigger than “Is 20x a better deal than 5x?”
The real multiplier is usage allowance × resets × long stretches of free time.
1. The short answer: 20x is 4 times 5x, and for people who max it out, a reset is worth 4 times more too
According to OpenAI, ChatGPT Pro comes in two tiers as of September 2026:
- Pro $100: 5x the Plus usage allowance
- Pro $200: 20x the Plus usage allowance
So Pro 20x gives you four times what Pro 5x does.[1]
Don’t read “20x” as “20 times 5x.” Both 5x and 20x are compared with Plus.
OpenAI’s official estimates for GPT-6 Astra in Work and Codex show the same ratio: 25–225 local messages per five hours on Pro 5x, and 100–900 on Pro 20x. Exactly four times the range. These are estimates, not fixed counts. How fast you burn through them depends on input and output length, reasoning settings, Fast mode and how complex the task is.[2]
A Full reset refreshes the eligible five-hour window and the weekly window.[2][3]
So the same “one reset” brings back more total usage on 20x than on 5x.
Same reset ticket. Bigger tank.
2. What does the “20” in 20x mean? It’s compared with Plus, so it’s 4x compared with 5x
Let’s do the math once, cleanly.
Call the Plus allowance for Work/Codex “1.”
- Plus: 1
- Pro 5x: 5
- Pro 20x: 20
Going from 5x to 20x adds 15 units. That is three more of your original 5, so the total becomes four times larger.
That “4x” matters a lot.
The name “20x” sounds huge on its own, but for someone already on 5x, the upgrade is 4x. On the other hand, if you keep running out of 5x, you are quadrupling the exact thing you are short of, and you will feel it.
If you don’t even use half of 5x each week, moving to 20x just gives you more unused allowance.
Compute you never use looks great on paper and is worth exactly $0 in practice.
3. Why does a Full reset feel so much bigger on 20x?
OpenAI says that when you use a saved Full reset, it refreshes the eligible five-hour and weekly windows and also moves your weekly reset date. A reset doesn’t permanently raise your allowance; it refills what you have at that moment.[2][3]
Let’s try a simplified thought experiment.
Say a 20x user fully burns through their allowance, then gets to use two separate Full resets, each one after running low.
normal 20
+ first reset 20
+ second reset 20
= 60
In Plus units, that looks like 60 units of work during that stretch.
Compare that with just one round of 5x with no resets and you get 60 ÷ 5 = 12x.
Cut out just that part and you want to shout:
“Buy 20x and it’s basically 60x, 12 times 5x!”
But that is not a fair plan comparison.
If the 5x user gets the same two Full resets and uses all of them too:
5 × 3 rounds = 15
20 × 3 rounds = 60
60 ÷ 15 = 4
The gap between the plans is still 4x.
So “12x” is just a theoretical number you get by comparing 20x plus several resets with a single, plain round of 5x in an unusual window. And because resets also move your weekly reset date, comparing by a fixed calendar week is misleading too.[2][3]
Still, for people who actually run out, one Full reset brings back a lot more on 20x. That part doesn’t change.
4. September 2026 really did feel like a week run by “The Reset Company”
The Astra launch had so many resets that you shouldn’t treat it as normal.
According to OpenAI’s help center, eligible existing Plus, Pro and Business users got one saved reset on September 3 and another on September 4, 2026. Separately, on September 7, eligible users got an automatic global reset.[2][3]
Then on September 9, OpenAI Status posted an incident where some Codex users saw “unexpected usage limit resets,” and reported it fixed the same day.[4]
After that, Thibault “Tibo” Sottiaux, who leads Codex at OpenAI, posted publicly that people whose banked reset didn’t fully apply to Work/Codex during the affected window would get another reset, plus an apology email.[5][6]
At that point, it feels like this from the user side:
“Tibo is printing resets again.”
Basically the Tibo Central Bank.
But there is an important line here. The help center calls these banked resets a promotional perk and clearly says future free resets are not guaranteed.[3]
So “I’ll buy 20x because I’ll get about two resets a week” is a risky way to plan a long-term subscription.
A launch party is a party. It is not a pension plan.
5. The more resets there are, the bigger the gap between heavy and light users
A reset isn’t worth “how many you got.” It is worth how much you had already used before it arrived.
If you still have 90% left and it goes back to 100%, you only gained 10%.
If you are at 0% and it jumps back to 100%, you basically get a whole extra round.
So the same gift looks like this:
- Someone with 5x to spare: nice, but it goes unused
- Someone who runs out of 5x every time: a big deal
- Someone who runs out of 20x: each refill is huge
This is where you can finally judge whether a pricier plan is worth it based on your actual usage, not the pricing page.
If you’re already maxing out the lower plan, you’ve proven you need more.
If you still have room left on the lower plan, “I might use it someday” is probably just FOMO.
6. In the end, the bottleneck isn’t the AI. It’s your free time
Even with four times the compute, nothing happens if the human handing out the work has no time.
You ask Astra to dig into something heavy.
You look at the results.
“This idea is weak. Let’s try this instead.”
You add data if needed.
You fix the broken setup.
You turn the steps that worked into a script.
All of that needs real, unbroken time, not scraps.
Which is why the AI era creates a weird reversal:
Having a powerful AI is becoming less rare than having the free time to keep giving it work.
If overtime eats one or two hours every day, you lose more than those hours. You also lose your recovery after work, the time it takes to get back into the problem, and the mental room to start a long experiment.
“I have 30 minutes today, so let’s build a new research setup” almost never happens.
Thirty minutes is too long for answering email and too short for building a research lab.
7. “Free time is powerful” is not just an excuse to be lazy
There is research that looks directly at free time and innovation.
Agrawal, Catalini and Goldfarb looked at how launches of innovative projects on a crowdfunding platform spike when colleges go on break. They suggest that free time helps not only with coming up with ideas, but also with the boring execution work, like admin, planning and promotion, that actually gets an innovation out into the market.[7]
That lines up surprisingly well with using AI.
When you build something new with AI, the flashy idea isn’t what eats your time. It’s boring steps like:
- setting up the environment
- getting the data ready
- writing tests
- keeping logs
- tidying up the README
- making it easy to run again
- fixing failure cases
Meanwhile, Pencavel’s research on working hours, using historical data from munitions factories, found that past a certain number of hours, each extra hour adds less and less output.[8]
Another analysis reported results consistent with the idea that long workweeks cause fatigue and too little recovery, which can hurt performance the following week.[9]
These studies weren’t about today’s AI researchers, so they don’t prove anything like “overtime always cuts creativity by X%.”
But the general direction, working longer doesn’t raise output in a straight line, and recovery and free time make you better at getting things done, matches how AI work actually feels.
8. If you get 20x for just one month, buy a “research factory,” not answers
If you’re using a big allowance for a short time, the biggest waste is firing off tons of questions and ending up with nothing but a pile of answers.
The smart move is the opposite:
Turn temporary compute into a setup you can keep using later.
OpenAI positions Work for long, multi-step research and finished deliverables, and Codex for software development. Codex can work with local folders, repositories, terminals and developer tools, and where it’s allowed, Work can also use local files and desktop apps.[10]
So for a one-month boost, aim for a setup that runs this loop without you giving instructions from scratch every time:
explore
↓
come up with a hypothesis
↓
build it
↓
run an experiment
↓
evaluate
↓
write down why it failed
↓
next hypothesis
“The AI researched 100 things for me” is fine.
“The AI built me a tool that tests 100 conditions automatically when I give it one candidate, and gives back a results table and failure logs”
is what’s still there next month, after the allowance runs out.
9. Have the AI build the setup too, and your free time turns into something you keep
If you’re building a research factory, code alone isn’t enough.
At the very least, you want:
- change history in Git
- pinned dependencies and versions
- a setup script
- run commands written in the README
- tests
- saved experiment settings and results
- failure logs
- fixed input and output formats
- the ability to reproduce it on a different computer
Do that, and the lab is still standing after your free time ends.
Even if you only have an hour after starting a new job or during a busy season, you don’t start with
“Today I’m setting up the environment from scratch.”
You start with
“Let me check what ran automatically up to yesterday.”
That is a big deal.
Permissions are a separate question, though. Keep high-risk, hard-to-undo things like API keys, secret keys, publishing to production, money transfers, real trades and deletions separate from your test environment, give only the access that’s needed, and keep a human check in place.
Having AI build your lab is powerful.
You don’t need to leave the vault key on the lab desk.
10. Is the extra $100 worth it? Think in hours saved, not “how many x”
Official pricing puts Pro 5x at $100 and Pro 20x at $200, so the difference is $100 a month. What you actually pay locally can vary with taxes, exchange rates and payment terms.[1]
If you see that $100 as just “AI money,” it feels expensive.
But if the goal is a focused month of building your setup, it’s better to ask how many hours of your own time it buys back.
- Save 10 hours: $10 per hour
- Save 20 hours: $5 per hour
- Save 40 hours: $2.50 per hour
This is not a promise of profit. It’s just simple division.
And setup work and automation aren’t a one-time time saver. If what you built keeps saving you an hour a week from next month on, the benefit adds up.
The decision gets pretty simple:
Leaning toward buying
- You often run out of 5x
- You can actually use up your banked resets
- You have a clear research or dev backlog you want done in a month
- You have a long stretch of free time right now
- You can leave behind setups and automation
Leaning toward skipping
- You have 5x left over
- Your main reason is “well, Astra just came out”
- You don’t know what you want it to do
- You don’t have time to review the results
- Your math only works if resets keep coming regularly
Count reset sprees as a bonus.
Don’t put them in your business plan as steady income.
11. Even the time you can’t upgrade costs you something
Here’s the interesting part: the value of an AI allowance changes depending on when you can use it.
Say someone wants to build a research setup all at once during a long vacation or a gap between jobs. They want to upgrade just for that period, but payment or plan-switching problems make them wait several days. The loss isn’t just the monthly price difference.
Your best free time and your biggest compute budget don’t line up.
That hurts.
Even if the same 20x is available next month, it’s not worth the same if you’re back at work every day.
AI pricing pages don’t list “your free time,” but it makes up a big part of what the subscription is really worth.
12. Bottom line: in the AI era, the scarce thing might be a big block of free time, not tokens
Pro 20x gives you four times the base allowance of Pro 5x.[1][2]
In unusual periods when Full resets pile up, the people who can max out their allowance get the most out of them. But there’s no guarantee how often free resets will come in the future.[3]
So “20x because there will be lots of resets” is a weak reason.
The strong reason is different:
You have a big block of free time right now, and you can use it to have Astra, Work and Codex build exploration tools, tests, environments, logs and automation.
If that’s you, beefing up your compute for just one month suddenly becomes a lot more worth it.
Free time you just spend is gone.
But if you use it to build something that keeps running even when you get busy, that free time pays you back again and again.
The strongest thing for AI-era R&D might not be an unlimited plan.
It might be having almost unlimited focus time, and using it to build a lab that runs without you.
And if you still burn through everything, the last request stays the same:
Tibo, lend me your computer. Unlimited, please.
References (10)
- OpenAI Help Center, “About ChatGPT Pro tiers help.openai.com
- OpenAI Help Center, “Managing usage with GPT-6 Astra in Work and Codex help.openai.com
- OpenAI Help Center, “How banked Codex resets work help.openai.com
- OpenAI Status, “Investigating unexpected usage limit resets status.openai.com
- Tibo Sottiaux, public X post on replacement resets for the affected window x.com
- OpenAI Forum, “Codex is for Everyone: Why Codex Matters Beyond Code forum.openai.com
- Agrawal, Catalini & Goldfarb, “Slack Time and Innovation,” NBER Working Paper 21134 nber.org
- John Pencavel, “The Productivity Of Working Hours,” Stanford SIEPR siepr.stanford.edu
- John Pencavel, “Recovery from work and the productivity of working hours,” Stanford SIEPR siepr.stanford.edu
- OpenAI Help Center, “ChatGPT Work and Codex help.openai.com
