ChatGPT Pro 200 usage cut: same price, half the plan

01 · What changed

The ChatGPT Pro 200 usage cut, in five lines

Pro 200 drops from 20x to 10x the Plus allowance for ChatGPT Work and Codex. The price stays at $200 a month (The Next Web).

GPT-6 Pro chat messages fall from 200 to 100 a week on the same plan.

A new Pro 500 plan costs $500 a month for 25x the Plus allowance and Ultrafast, a speed tier that generates tokens up to 8x faster in Codex (Neowin).

The cut lands on 30 October 2026. Eligible existing subscribers keep 20x through 29 October, as long as their Pro 200 subscription stays active.

Existing subscribers get a one time credit grant, reported as 62,500 credits, worth about $2,500 of usage, expiring 31 December 2026 (Hardware Busters). The first reports said the amount was not disclosed, so check what actually landed in your account.

02 · Timeline

Taken off the shelf, put back smaller

Date	What happened
10 September	OpenAI pauses new Pro 200 sign ups and upgrades. Reports tie it to GPT-6 Astra demand straining capacity. The announcement gives no reason.
29 September	DevDay. Pro 200 reopens at the same price with 10x instead of 20x. Pro 500 launches.
29 October	Last day of the old 20x allowance for eligible existing subscribers.
30 October	The cut applies to everyone on Pro 200.
31 December	The one time credit grant expires.

That sequence is why people call it a rug pull rather than a repricing. The plan disappeared for nineteen days, during a capacity crunch, then came back on the biggest stage of OpenAI's year with half the allowance, while everyone was looking at the new launches. Nothing about it is hidden, exactly. It just arrived in the one week when nobody was reading the small print.

03 · The math

The volume discount is gone

The cleanest way to see the change is to price one unit of usage, where one unit is a Plus plan's worth of allowance. Divide each plan's price by its multiplier:

Plan	Price per month	Allowance	Cost per 1x Plus
Plus	$20	1x	$20
Pro 100	$100	5x	$20
Pro 200, until 29 October	$200	20x	$10
Pro 200, from 30 October	$200	10x	$20
Pro 500	$500	25x	$20

Multipliers as reported by paddo.dev, Kingy AI and Hardware Busters on 29 and 30 September 2026.

Until now Pro 200 was the only plan with a bulk discount. It bought twice as much per dollar as every other tier, which is exactly why the heaviest Codex users chose it over the API. After 30 October every tier costs the same $20 per unit. Pro 200 becomes ten Plus subscriptions in one invoice.

For someone who used the whole 20x, that is a 100% effective price increase. And Pro 500 is not the escape hatch it looks like: it also costs $20 per unit. What the extra $300 buys you is speed, not a better rate. As one analysis put it, "Paying more used to get you a better rate."

04 · OpenAI's side

The argument for the cut, and where it runs out

OpenAI does have a case, and it is worth hearing before you decide anything. Thibault Sottiaux, who leads Codex and ChatGPT Work, said the new allowance is roughly half the old one when valued at API list prices (Xenospectrum). Three things are meant to soften it.

The models got cheaper. GPT-6.1 Sol launched at $2 input and $10 output per million tokens. If your allowance is counted in API dollars, a cheaper model stretches the same dollars further.

The five hour cap stays gone. The new Pro 200 does not bring back the 5-hour limit, so you can spend a week's allowance in a few long sessions instead of rationing it by the hour.

You will get more done anyway. OpenAI's line is that you can still do more with Pro than you could before, because the new models are better per token.

Each of those is partly true, and none of them answers the actual complaint.

Cheaper models are the normal course of things. Every generation gets more efficient. Pricing used to pass that gain on to you. Now it pays for a cut.

It only works if you change models. If your work needs Astra, the model whose demand caused the capacity crunch in the first place, Sol being cheap does not help you. As Xenospectrum put it, the real impact "can't be determined from the API price list alone."

The credit grant is temporary. The cut is not. A one time grant that expires on 31 December softens one quarter. The 10x allowance is the plan from now on.

05 · Who it hits

The customers OpenAI courted hardest

Heavy Codex users

People running long agentic sessions every day, who chose Pro 200 precisely because the per unit rate beat the API. They lose the rate and keep the workload.

Astra workloads

If your tasks need the top model and fall apart on Sol, the cheaper model argument does nothing for you. Your allowance simply halves.

Solo developers and small studios

Anyone who budgeted the year around a fixed 20x allowance now has to budget for either half the work or twice the spend.

If you used less than half your allowance, you may never notice. If you regularly hit the ceiling, the same work now costs twice as much.

06 · Before 30 October

Six things to do in the next four weeks

1. Measure what you actually use

Look at a normal month, not a guess. If you used under half of 20x, the new Pro 200 still fits and you can stop reading here. If you are a Termdeck user, the usage page already shows the API equivalent cost of your Codex and Claude turns, which is the same yardstick OpenAI used to set the new allowance.

2. Test GPT-6.1 Sol on your real work

OpenAI's whole argument rests on Sol. Run it on your actual tasks, not a demo. One independent test found it fixed 94% of new tests but broke existing tests in 5 of 20 runs (paddo.dev). Know which of those you are.

3. Spend the credit grant on purpose

If the one time credits landed in your account, aim them at your most expensive work before 31 December rather than letting them leak into everyday chats.

4. Price the alternatives on one task

Every OpenAI tier now costs $20 per unit, so Pro 200 has no price advantage over Plus, Pro 100 or the API. Run the same task on a second engine and compare the result, not the marketing.

5. Say so in writing

Subscribers are asking OpenAI to grandfather the 20x allowance under #KeepPro20x. If you are in the EU, Hardware Busters points to the rules on contract changes in Directive (EU) 2019/770. Read your terms before relying on it.

6. Decide before renewal

If the plan no longer pays for itself, downgrade before your billing date. Do not let auto renew make the decision for you.

07 · The real lesson

Never let one vendor's pricing own your workflow

This is bigger than OpenAI. A flat rate AI subscription is a subsidy, and a subsidy lasts until it stops being worth paying for. Vendors use generous plans to win you, then pull them back once you depend on them. The Pro 200 discount lasted exactly until demand caught up with capacity. Every other generous plan, from every vendor, sits on the same curve.

The way to protect yourself is not loyalty to the cheapest vendor this month. It is making a switch cost you one click instead of one month. In practice that means three things.

Your history lives in files you own. Claude Code and Codex both write every session to disk on your machine. If your workflow depends on a chat you can only see in one vendor's web app, you cannot take it anywhere.

A second engine is already installed and signed in. The day a plan changes is the wrong day to learn a new CLI.

You can see what each plan actually buys. Not the multiplier on the pricing page: the tokens your own work burns, per model, per day.

That is the setup Termdeck is built around, and I would say this even if it were not. Termdeck drives the official Claude Code, Codex and Grok CLIs on your own machines, unmodified and under your own subscriptions, and you pick the engine per chat. Your accounts panel keeps a Claude login and a ChatGPT login on the same machine, shows the live rate limits each provider reports, and switches accounts in one click. When Codex hits its weekly wall on a Tuesday, your next chat goes to Claude Code without changing tools. The Claude Code and Codex in one UI guide walks through the setup.

Termdeck is independent and has no affiliation with OpenAI, Anthropic or xAI. We drive their CLIs; we do not resell their plans.

Honest limits

What Termdeck does not fix here

It does not make Pro 200 cheaper. Your turns are billed against your own ChatGPT plan, exactly as they are in a terminal. A 10x allowance is a 10x allowance wherever you run Codex.

It does not pool limits. Switching accounts moves you to a different plan's allowance. It does not merge two allowances into one, and it does not get around anyone's terms.

Its cost figures are estimates. The usage page applies published API rates to the tokens your transcripts recorded. That is the right yardstick for this decision, but it is not your OpenAI invoice.

It costs money too. Termdeck is a paid product with a 14-day free trial. If all you want is to stop paying OpenAI $200, downgrading is free.

FAQ

The Pro 200 cut, answered

Did ChatGPT Pro go up in price?

Not on paper. Pro 200 still costs $200 a month, but from 30 October 2026 it includes 10x the Plus allowance instead of 20x. You pay the same for half the usage, so the effective price per unit of work doubles.

When does the ChatGPT Pro 200 usage cut take effect?

On 30 October 2026. Eligible existing subscribers keep the old 20x allowance through 29 October as long as their Pro 200 subscription stays active. New subscribers get 10x from the start.

What is ChatGPT Pro 500?

A $500 a month plan announced at OpenAI DevDay 2026, with 25x the Plus allowance and access to Ultrafast, which generates tokens up to 8x faster. Per unit of usage it costs the same as Plus, so you are paying for speed rather than a better rate.

Do existing Pro subscribers get anything for the cut?

A one time credit grant, reported as 62,500 credits worth about $2,500 and expiring on 31 December 2026. Early reports said the amount was undisclosed, so check your own account for what you received.

Why did OpenAI pause Pro 200 sign ups?

OpenAI paused new Pro 200 sign ups on 10 September 2026 without giving a reason. Reports tie the pause to infrastructure strain from GPT-6 Astra demand. The plan reopened at DevDay on 29 September with the smaller allowance.

Can I use Claude Code and Codex together?

Yes. Both are command line tools that can sit on the same machine, each signed in to its own plan. Termdeck runs both from one browser tab and lets you pick the engine per chat, so a limit on one plan does not stop your work.

Sources, all accessed 1 October 2026: The Next Web, Neowin, CNBC, paddo.dev, Hardware Busters, Xenospectrum, Kingy AI. OpenAI's plans move quickly; check its own pricing page before you act on any number here.

Get started

Keep a second engine one click away.

Run Claude Code, Codex and Grok from one browser tab, each under your own plan, and see what every model actually costs you. 14-day free trial, and no charge if you cancel before day 15.

Set up both engines Start free trial

OpenAI kept ChatGPT Pro at $200 and halved what it buys.