ChatGPT Pro Max Is Pro 500: $500, 25x, and a $200 Plan Cut
ChatGPT Pro Max is real, but it launched under a different name: Pro 500, a $500 a month plan with 25x Plus usage and exclusive access to GPT-6 Astra Ultrafast, announced at OpenAI DevDay on September 29, 2026. The same day OpenAI reopened the $200 Pro plan to new buyers and cut its allowance from 20x to 10x Plus, so every ChatGPT tier now costs the same $20 per unit of usage.
Key takeaways
- ChatGPT Pro Max shipped as "Pro 500": $500 a month, 25x Plus usage, and Ultrafast in Codex and ChatGPT Work. The leak got the price and the speed pitch right. The name changed, and the allowance is 25x, not the 20x we modelled.
- The $200 Pro plan is open again, at half the value. OpenAI's Codex lead Thibault Sottiaux said it will "net out at half the dollar in API spend compared to the old Pro $200 plan". In practice that is 10x Plus instead of 20x.
- Existing $200 subscribers keep 20x until October 29. From October 30 they drop to 10x, and they get a one-time grant of 62,500 credits (about $2,500) that expires December 31, 2026, according to OpenAI's help center as quoted by TNW and The Decoder.
- The volume discount is gone. Plus, Pro 100, Pro 200 and Pro 500 all price out at $20 per unit of Plus-equivalent usage. The old $200 plan was $10. Buying a bigger plan now gets you features and speed, not cheaper work.
- No 5-hour window on Pro. Sottiaux committed to "not reintroducing the 5h limit", so Pro usage is a weekly budget you can spend in one sitting. Plus keeps its 5-hour window.
- Ultrafast is expensive to burn. GPT-6 Astra Ultrafast lists at $60 input and $300 output per million tokens in the API, 6x standard Astra. If the subscription meters it at that rate, a 25x plan buys roughly 4x Plus worth of Ultrafast work.
- Plus did not change. The viral "$20 is now $10" line is a misreading; no OpenAI document or report supports it.
Updated September 30, 2026, after DevDay. This page first ran on September 25 as a write-up of the leaked $500 plan. We rewrote it once OpenAI announced Pro 500 and the $200 plan change, kept the original leak scorecard, and graded it against what shipped. Primary sources: OpenAI on X and Thibault Sottiaux on X; plan details via TNW, The Decoder and Engadget.
What OpenAI announced on September 29
Two things landed within a few hours of each other. First, the night before DevDay, Sottiaux posted that the $200 Pro plan, closed to new buyers since September 10, would reopen the next day with a new usage formula. His opening was blunt: "In effect, if you do the math, it will net out at half the dollar in API spend compared to the old Pro $200 plan" (full post).
Second, at DevDay itself, OpenAI launched the plan that leaked a week earlier. From the official announcement: "Ultrafast is available today for GPT-6 Astra in Codex, ChatGPT Work, and the API, with GPT-6.1 Sol coming soon. To access it in Codex and ChatGPT Work, we're introducing Pro 500, a new plan with our highest usage limits (25x Plus) and access to Ultrafast."
So the thing people searched for as ChatGPT Pro Max is Pro 500. The $500 price, the focus on Codex and Work, and the speed angle all held. What the leak could not tell us was the allowance, and it is the allowance that decides whether any of these plans is worth the money.
The new ChatGPT plans and limits, side by side
Here is the ladder as of September 30. Usage is expressed the way OpenAI now expresses it, as a multiple of Plus. The per-unit column is the price divided by that multiple.
Sources: OpenAI, TNW, Mixed. Reported by the same outlets: GPT-6 Pro chat messages on the $200 plan fall from 200 to 100 a week on the same schedule, and keeping the old allowance until October 29 "does not upgrade your plan or add Ultrafast". OpenAI's help center page on Pro tiers is the canonical source; we could only read it through press quotes.
The math: every tier now costs $20 a unit
Our September 25 version of this page argued that a $500 plan at the old 20x allowance would cost $25 per unit against $10 on the $200 plan, and that OpenAI would need to reach 50x Plus to match the $200 plan's rate. OpenAI solved that gap from the other end. It did not make the top plan cheaper per unit. It made the $200 plan as expensive per unit as everything else.
That is the real change, and it is bigger than the headline number. Until this week the ChatGPT ladder, like Claude's and Google's, rewarded heavy users with a volume discount: $20 bought 1x, $200 bought 20x. Now it is flat. $20, $100, $200 and $500 buy 1x, 5x, 10x and 25x. Whatever plan you pick, a unit of work costs the same. Moving up buys you room, GPT-6 Pro, and on the top plan, speed.
Sottiaux gave the reasoning in the same post. OpenAI does not want "to artificially inflate the API list prices to make it look like you are getting a lot", and it expects API prices to fall far enough that "it makes sense for most to buy usage as needed". Read plainly, subscriptions are being tied to API list prices, and the subsidy on the $200 tier is being wound down. Harrison Kinsley (Sentdex) put the same thing in one line on X: "the api prices are the real prices and the gap will continue to close."
What Ultrafast costs inside a 25x allowance
Ultrafast is GPT-6 Astra served at up to 300 tokens per second, which OpenAI says is up to 8x faster in Codex. In the API it lists at $60 per million input tokens and $300 per million output tokens, against $10 and $50 for standard Astra (VentureBeat). That is exactly 6x.
OpenAI has not published how Ultrafast draws on a Pro 500 allowance. But Sottiaux describes the whole system in API-dollar terms, and if Ultrafast is metered at its API rate, the 25x allowance holds about 25 / 6, roughly 4x Plus worth of Ultrafast work. That is our estimate, not an OpenAI number. It fits one widely shared post from a heavy Codex user who said they emptied a whole week of Pro 500 on Astra Ultrafast in under 30 minutes. We could not verify that account, but the arithmetic does not make it implausible.
The practical rule: on Pro 500, switch Ultrafast on for the runs where you are actually waiting, and leave long background Codex jobs on standard speed. Otherwise you are paying 6x for latency nobody is watching.
Where the speed comes from
OpenAI signed a deal worth more than $10 billion with Cerebras in January 2026, and in August Cerebras confirmed it serves GPT-5.6 Sol Ultrafast. Astra Ultrafast is reported to run on the same hardware, but OpenAI's own materials do not name a vendor for the Astra tier. Treat Cerebras as likely, not confirmed.
The leak scorecard, graded against what shipped
We publish these scorecards because a leak is only useful if you know which parts to trust. The Opus 5.5 leak was right on eight of nine claims. This one did about as well.
Our own call deserves a grade too. We said the allowance was unknown and modelled 20x; it is 25x. We said OpenAI would need about 50x to match the old $200 rate. It never tried, because it moved the $200 plan to $20 a unit instead.
Why OpenAI cut the $200 plan now
The capacity story from earlier this month still frames it. GPT-6 Astra launched September 3, demand outran compute, and on September 10 OpenAI paused new $200 Pro sign-ups; Sottiaux said those subscriptions "put the most strain on our systems" (Fortune). A plan that was closed because it cost too much to serve could only come back cheaper to serve. It did.
The other half is API pricing. The week before DevDay, OpenAI shipped GPT-6 Sol and GPT-6 Luna at half the list price of the GPT-5.6 models they replace; Sol went from $4 and $20 to $2 and $10 per million input and output tokens (MarkTechPost). At DevDay it added GPT-6.1 Sol at the same $2 and $10, which OpenAI pitches as close to Astra on coding and computer use at a fifth of the price (VentureBeat). Sottiaux's argument is that cheaper models mean you "still get more work done than if you were on the Pro $200 subscription one month ago", as long as you work on Sol rather than Astra.
That argument is true in one direction and not the other. If your usage is on Sol, halved prices roughly offset a halved dollar budget. If you bought the $200 plan for Astra, whose price did not fall, you get about half as much Astra as you did in August.
What the $200 plan still gets that the others do not
Sottiaux promised two things in exchange. No 5-hour window, so the weekly budget can be spent in one long session. And "more things to the subscription that won't draw on the usage". The first of those arrived at DevDay: Dots, OpenAI's always-on personal agents, which run on their own cloud computer and browser and connect to thousands of apps. The first Dot is included with Pro plans, and Dot conversations do not count against your usage (Decrypt, Engadget).
The reaction, and what is fair in it
X did not take it well. "Ooooof. Mad respect for the transparency here but this hurts a lot," wrote Theo Browne. Others called it a rug pull, or recast the plans as "$200 is now $100". Some of that is fair: the change landed the night before a launch event, and existing subscribers find out their allowance halves in a month. The "$20 is now $10" version is not fair. Nothing changed on Plus.
The more measured take came from dax (thdxr), who pointed out that these subscriptions "at best break even and eat up limited compute" at both OpenAI and Anthropic. Credit where due: OpenAI stated the cut as a number before anyone had to reverse-engineer it. Anthropic's limit changes have mostly been found by users first.
ChatGPT Pro vs Claude Max after the cut
On paper, Anthropic now has the cheaper heavy-use plan. Claude Max 20x is still $200 for 20x Claude Pro, a volume discount OpenAI just dropped. Two caveats keep it from being a simple win.
First, a Plus unit and a Claude Pro unit are not the same amount of work, and neither company publishes the underlying numbers, so "20x vs 10x" compares two opaque multipliers. Second, Claude Max still has 5-hour sessions, which cap how hard you can push in one sitting. For the model side of the choice, see our running leaderboard of coding models; for the agents, Codex vs Claude Code vs Cursor.
Should you buy Pro 200, Pro 500, or neither
- You hit the Plus 5-hour window but not the weekly cap: Pro 100. Same per-unit price, no 5-hour window.
- You work on GPT-6 Sol or 6.1 Sol most of the day: Pro 200 is fine. Sol's price cut roughly covers the smaller budget.
- You live on Astra in Codex: price Pro 200 against the API before renewing. With no discount left in the plan, pay-as-you-go may cost less in a light month.
- You sit and watch long Codex runs: Pro 500, with Ultrafast used on purpose. At $100 an hour of engineer time, the $300 over Pro 200 pays back if it saves three hours of waiting a month.
- You are on the old $200 plan: nothing changes until October 30. Use the 62,500 credits before December 31, then decide on real usage data, not on the 20x you used to have.
For teams, the lesson from this month is dependency. In four weeks one vendor launched a flagship, cut limits, closed its top plan, opened a $500 plan and halved the $200 one. None of it was malicious, and all of it changed what a seat costs. If your workflow can run an OpenAI model inside Claude Code or switch agents without a rewrite, a pricing change is a config edit. Fast agents also fail fast, as the agent-driven OpenAI breach showed earlier this month, and Ultrafast shortens the time before a human looks.
Stop paying for AI seats you cannot measure
ChatGPT seats now run from $20 to $500, the $200 allowance just halved, and every tier is priced off API list rates. The number that matters is cost per finished task on your own work, and whether a seat or the API is cheaper for it.
Valletta Software Development sets up the layer that tells you: provider-agnostic routing between OpenAI, Anthropic and open models, per-task cost and latency telemetry, and a workflow that survives the next pricing change. Tell us what your team runs and what it spends.
FAQ
What is ChatGPT Pro Max?
ChatGPT Pro Max was the leaked name of OpenAI's $500 a month plan. It launched on September 29, 2026 as Pro 500, with 25x Plus usage and exclusive access to GPT-6 Astra Ultrafast in Codex and ChatGPT Work.
How much is ChatGPT Pro 500?
$500 a month in the US, or $6,000 a year before tax. That buys 25x Plus usage, the same $20 per unit as every other ChatGPT tier.
What are the ChatGPT Pro usage limits now?
Plus is 1x with a 5-hour window, Pro 100 is 5x, Pro 200 is 10x for new buyers, and Pro 500 is 25x. Pro plans have no 5-hour window, only a weekly limit. Existing Pro 200 subscribers keep 20x until October 29, 2026.
Did OpenAI cut the $200 ChatGPT Pro plan in half?
Yes, in allowance. OpenAI said the reopened $200 plan nets out at half the API-dollar value of the old one, which is 10x Plus instead of 20x. New buyers get 10x from September 30; existing subscribers move to 10x on October 30 and get a one-time 62,500-credit grant.
Is there a ChatGPT 50x plan?
No. The highest ChatGPT plan is Pro 500 at 25x Plus. The 50x figure came from our earlier estimate of the allowance a $500 plan would need to match the old $200 plan's price per unit.
Did ChatGPT Plus get cut to $10 worth of usage?
No. Nothing changed on Plus. The claim spread on X as a misreading of the Pro 200 change.
What is GPT-6 Astra Ultrafast?
A faster serving tier of GPT-6 Astra, up to 300 tokens per second and up to 8x faster in Codex. In the API it costs $60 input and $300 output per million tokens, 6x standard Astra. In Codex and ChatGPT Work it is only on Pro 500 and Enterprise.
Is ChatGPT Pro 500 worth it?
Only if waiting is your bottleneck. It costs the same per unit as the cheaper plans, so you pay the extra $300 over Pro 200 for more room and Ultrafast. If Ultrafast is metered at its 6x API rate, heavy use drains the allowance quickly.
ChatGPT Pro or Claude Max: which is better value now?
On the published multipliers, Claude Max 20x at $200 keeps a volume discount ChatGPT dropped. But the units are not comparable across vendors, Claude Max has 5-hour sessions, and only ChatGPT Pro 500 offers a fast serving tier.