ChatGPT Pro $200 Is Back, With Half the Usage

The $200 ChatGPT Pro plan is back. After three weeks of paused sign-ups, OpenAI reopened it today, September 30.

On the surface that's good news. The plan everyone was locked out of is available again, it has no five-hour limit, and it comes with some shiny new extras from DevDay.

Look at what it actually buys, though, and it's a different plan. It now gives you half the usage it did before, at the same $200. The Codex and ChatGPT Work allowance drops from 20 times what Plus gets to 10 times. GPT-6 Pro messages in ChatGPT drop from 200 a week to 100.

That turns the $200 tier from the best-value AI subscription on the market into one with no volume discount at all. And it happens at the worst possible moment for OpenAI, because for the same $200, Claude Max 20x still gives you 20 times its $20 plan, and Anthropic has been raising its limits this year, not cutting them.

ChatGPT Pro $200, before and after

Codex and Work usage, vs. Plus

20x → 10x

Halved for new subscribers now, everyone else on October 30

What 1x of Plus usage costs

$10 → $20

The same as on Plus itself: no bulk discount left

GPT-6 Pro messages a week

200 → 100

In ChatGPT, on the same $200 plan

Claude Max 20x, for comparison

$10

Per 1x of Claude Pro, with limits raised twice this year

Sign-ups for Pro $200 were paused from September 10 and reopen today, September 30, at the new rate. Existing subscribers keep the old quota until October 29 and get a one-time $2,500 usage credit that expires at the end of 2026. Sources: OpenAI, Anthropic.

Six weeks, from "exactly 20x" to half of that

To understand why people are annoyed, you need the timeline. It moved fast.

• August 30. Thibault Sottiaux ("Tibo"), who leads engineering for Codex and ChatGPT Work at OpenAI, confirmed that Pro $200's 20x figure "does exactly what it says on the tin," according to Kingy AI's timeline.
• September 3. GPT-6 Astra launched at $10 per million input tokens and $50 per million output tokens.
• September 10. OpenAI paused new Pro $200 sign-ups. "Demand for Astra is really unprecedented," Tibo told TechCrunch. "I've not seen anything like it until now." Pro subscriptions, OpenAI said, put the most strain on its systems. Fortune reported the same the next day.
• September 22. OpenAI launched GPT-6 Sol and GPT-6 Luna at half the price of the previous generation. The same day, Anthropic shipped Claude Opus 5.5 and raised the five-hour limits on Pro, Max and Team.
• September 28–29. Tibo announced on X that Pro $200 would come back with a new usage calculation, and at DevDay OpenAI confirmed the numbers: 10x Plus instead of 20x, and a new $500 plan on top.
• September 30 (today). Sign-ups reopen at the new rate.
• October 30. Existing subscribers move to the lower quota. Until October 29 they keep the old one, plus a one-time $2,500 usage credit that expires at the end of the year.

A month ago the 20x plan did "exactly what it says on the tin." Now the tin says 10x.

What OpenAI actually said

To OpenAI's credit, it didn't hide the cut. Tibo put it plainly:

The argument OpenAI is making is that this is fine because models got cheaper. Engadget and Android Headlines both carry the same line from Tibo: cheaper API prices for models like GPT-6 Sol mean you can still get plenty of work done with a smaller allowance. He also said subscribers will "get more work done" than on the old plan.

Tibo made four promises alongside the change: no five-hour cap, future API price cuts passed through to subscriptions, no inflated list prices behind the subscription discount, and some "unmetered extras" he didn't name. OpenAI still hasn't said what the allowance is in dollars, only that it's half of what it was.

Some of that is true. I'll come back to it. But look at what the change does to the plan's value first.

The best-value plan now has no discount at all

Here's the thing that bothers me most.

Before today, the $200 plan was the only ChatGPT plan where paying more got you more per dollar. Plus is $20 for 1x. Pro $100 is $100 for 5x, so $20 per 1x, no better than Plus. Pro $200 was $200 for 20x: $10 per 1x, half price. That was the whole reason heavy Codex users bought it.

After the change, it's $200 for 10x. That's $20 per 1x, exactly the same as Plus. The new $500 plan is 25x, also $20 per 1x.

• Until October 29
• From October 30

Plus

$20

$20

Pro $100

$20

$20

Pro $200

$10

$20doubled

Pro $500

$20

Sources: OpenAI’s DevDay 2026 announcements and the OpenAI Help Center article on ChatGPT Pro tiers, as reported by The Next Web, Engadget and BGR.

Show the data

Plan	Price a month	Usage before	Usage after	Per 1x before	Per 1x after
Plus	$20	1x	1x	$20	$20
Pro $100	$100	5x	5x	$20	$20
Pro $200	$200	20x	10x	$10	$20
Pro $500	$500	didn't exist	25x	–	$20

So OpenAI went from having the best subscription deal in AI coding to having no bulk discount anywhere in its lineup. Every plan now charges the same rate. The more expensive plans just let you buy more of it at once.

Put the plans on one chart and it's even clearer. ChatGPT's plans now sit on a perfectly straight line. Claude's $200 plan still bends upward:

• ChatGPT, from October 30
• Claude (Anthropic)

Usage, as a multiple of the $20 plan

0x5x10x15x20x25x

ChatGPT, from October 30Claude (Anthropic)

$0$100$200$300$400$500

Price a month

Sources: OpenAI’s DevDay 2026 announcements and the OpenAI Help Center article on ChatGPT Pro tiers, as reported by The Next Web, Engadget and BGR. Anthropic’s tiers are from its Max plan help article. Each company measures against its own $20 plan, so a multiple on one isn’t the same amount of work as on the other.

Show the data

Plans	Plan	Price a month	Usage vs. the $20 plan
ChatGPT, until October 29	Plus	$20	1x
ChatGPT, until October 29	Pro $100	$100	5x
ChatGPT, until October 29	Pro $200	$200	20x
ChatGPT, from October 30	Plus	$20	1x
ChatGPT, from October 30	Pro $100	$100	5x
ChatGPT, from October 30	Pro $200	$200	10x
ChatGPT, from October 30	Pro $500	$500	25x
Claude (Anthropic)	Pro	$20	1x
Claude (Anthropic)	Max 5x	$100	5x
Claude (Anthropic)	Max 20x	$200	20x

If you're a heavy user, that bend is the only part of the chart that matters.

Meanwhile, $200 still buys 20x at Anthropic

This is why the change makes so little sense right now. OpenAI isn't cutting in a vacuum. The obvious alternative costs the same and didn't get cut.

Claude Max 20x is $200 a month for 20 times the usage of Claude Pro. And Anthropic has been moving in the other direction all year:

• May 6. Anthropic doubled Claude Code's five-hour limits on Pro, Max, Team and seat-based Enterprise, and removed the peak-hours reduction for Pro and Max. The capacity came from a deal for all of SpaceX's Colossus 1 data center: more than 300 megawatts and over 220,000 NVIDIA GPUs.
• September 22. With Opus 5.5, Anthropic raised the five-hour limits again and gave every subscriber a rate-limit reset they can bank and use whenever they want, until October 22.

• Until October 29
• From October 30
• Claude (Anthropic)

ChatGPT Pro $200, until October 29

20x

ChatGPT Pro $200, from October 30

10x−50%

Claude Max 20x, today

20x

Sources: OpenAI’s DevDay 2026 announcements and the OpenAI Help Center article on ChatGPT Pro tiers, as reported by The Next Web, Engadget and BGR. Claude: Anthropic’s Max plan help article. The two companies’ $20 plans aren’t the same size, so compare the direction of travel, not the exact amounts.

Show the data

Plan	When	Usage vs. the $20 plan
ChatGPT Pro $200	until October 29	20x
ChatGPT Pro $200	from October 30	10x
Claude Max 20x	today	20x

I want to be fair here: Claude Max isn't literally unlimited. It has a five-hour window and a weekly cap across all models, and Anthropic doesn't publish exact token numbers. The two "20x" figures also aren't the same amount of work, because each is measured against a different $20 plan. But in practice, on Max 20x I rarely think about limits at all. It's the closest thing to "unlimited" you can buy for $200.

A big reason is cache reads. In a long Claude Code session, most of what the model processes is the same codebase and conversation being re-read turn after turn. On the API, that's metered. On Max, it's part of the flat rate. I wrote about how much cache pricing matters in my Sonnet 5.5 review, and it's exactly why a flat subscription is worth so much more than its sticker price for agent work.

Anthropic also cut Opus 5.5's API price to $4 / $20, down from $5 / $25 on Opus 5, and cached reads to $0.20. So Claude got cheaper per token and more generous per subscription in the same month OpenAI halved its plan. Tibo himself agreed with a critic on X that OpenAI should "simply beat Opus 5.5 on value per dollar," Kingy AI reported. This change does the opposite.

Does "the models got cheaper" hold up?

Partly, and only for some models.

The new allowance is worth half as many API dollars. So how many tokens you get depends on whether the model you use also got cheaper. I did the arithmetic for each one:

GPT-6 Sol

100%

GPT-6 Luna, output

120%

GPT-6 Luna, input

100%

GPT-6 Astra

50%

GPT-5.6 Sol

50%

My arithmetic from OpenAI’s stated rule (half the API-dollar value of the old plan) and its API prices per million tokens, before and after the September 22 price cuts. OpenAI hasn’t published the dollar amount itself, only the ratio.

Show the data

Model	API price per 1M tokens	Tokens vs. the old Pro $200
GPT-6 Sol	$2 / $10, down from GPT-5.6 Sol's $4 / $20	100%
GPT-6 Luna, output	$0.50, down from $1.20	120%
GPT-6 Luna, input	$0.10, down from $0.20	100%
GPT-6 Astra	$10 / $50, unchanged	50%
GPT-5.6 Sol	$4 / $20, still on sale	50%

• GPT-6 Sol costs half what GPT-5.6 Sol did ($2 / $10 instead of $4 / $20), so half the dollars buys about the same number of Sol tokens. For Sol-only users, the cut roughly cancels out.
• GPT-6 Luna got cheaper by more than half on output, so Luna users actually get about 20% more output tokens.
• GPT-6 Astra didn't get cheaper. It's still $10 / $50. Astra users get half as much. And Astra is the model whose demand caused the pause in the first place. It's the reason most people wanted Pro $200.
• GPT-5.6 Sol is still on sale at the old price, so anyone who stayed on it also gets half.

So OpenAI's defence works if you move your work down to the cheaper model. If you paid $200 for the frontier model, which is the point of a $200 plan, you're getting half.

One thing that's genuinely better: there's no five-hour limit on Pro $200, and Tibo promised it won't come back. If your usage comes in bursts, that's worth something. Existing subscribers also get a $2,500 credit to soften the switch. Neither of those changes what the plan buys from October 30 onwards.

Why I think OpenAI did it

This isn't a mystery. OpenAI ran out of capacity. Astra demand was "unprecedented," and heavy Pro users were costing the most to serve, so OpenAI closed sign-ups. Cutting the allowance in half and adding a $500 tier with exclusive Ultrafast speed (up to 300 tokens per second in Codex, according to OpenAI) is how you reopen sign-ups without buying twice as many GPUs.

That's a reasonable business decision. It's just a bad deal for the people who were paying $200, and badly timed. Two weeks ago, a heavy Codex user had one good reason to stay: Pro $200 was the best value. Now that reason is gone, and Claude Max 20x is sitting right there at the same price.

It also connects to something I wrote in the summer. In my post on Fable 5 and Sonnet 5, I called it the junior engineer problem: a cheap model isn't cheap if it burns ten times as many tokens getting the job done. The same logic applies here in reverse. A $200 plan isn't a deal because it costs $200. It's a deal because of how much work it gets through before you hit the wall. OpenAI just moved the wall.

What I'd do if you're on Pro $200

• If you mainly use GPT-6 Sol: you probably won't notice much. The price cut covers the allowance cut.
• If you rely on GPT-6 Astra or GPT-6 Pro: you're losing half. Use the old quota and the $2,500 credit until October 29, and use that time to decide.
• If you mostly use Codex for heavy agentic coding: try Claude Max 20x with Claude Code for a month. It's the same $200, still 20x, and the limits have only gone up this year.
• Don't upgrade to Pro $500 just for the usage. It costs $20 per 1x of Plus, the same rate as every other plan. The extra $300 buys Ultrafast speed, not better value.

The verdict

"Pro $200 is back" sounds like good news. It isn't, really. OpenAI brought back the name and the price, but not the plan. Pro $200 used to be the one ChatGPT plan with a volume discount: half price per unit of usage. From October 30, it has none.

I understand why OpenAI did it. Capacity is finite and Astra is expensive to run. But the timing couldn't be worse. Anthropic cut Opus prices, raised its limits twice this year, and still sells 20x for $200. OpenAI now sells 10x for the same money and asks you to believe that cheaper models make up for it. For Sol, they do. For Astra, the model you actually paid $200 for, they don't.

The best subscription in AI coding used to have OpenAI's name on it. Today, it doesn't.

---

A note on sources: plan changes, dates and the $2,500 credit come from OpenAI's DevDay 2026 announcements and the OpenAI Help Center article on ChatGPT Pro tiers, as reported by The Next Web, Engadget, BGR and Unite.AI. The pause is from TechCrunch and Fortune. Model prices and Tibo's August 30 remark are from Kingy AI. Claude's plans and limits come from Anthropic's Max plan article, its May 6 announcement and the Opus 5.5 launch. The per-model token chart is my own arithmetic from OpenAI's stated rule and its API prices. Everything written as "I think" is my opinion.