ChatGPT Pro Max: The $500 Plan That Sells Speed, Not Quota
ChatGPT Pro Max is an unreleased $500 per month ChatGPT plan that surfaced in OpenAI's front-end code on September 24, 2026, and the only thing its leaked description adds over the $200 Pro plan is "Fastest Work and Codex". OpenAI has not announced it, and nothing in the leak promises more usage, which makes it the first ChatGPT tier that charges more per unit of usage than the tier below it.
Key takeaways
- Price: $500 a month, leaked, not announced. Strings in ChatGPT's front-end configuration, found by Tibor Blaho and TestingCatalog, name a "Pro Max" plan at $500. That is 2.5x the $200 Pro plan and 25x Plus.
- The only feature line that differs is "Fastest Work and Codex". Higher usage limits are not in the leaked text. TestingCatalog itself says they are possible but unconfirmed.
- If Pro Max carries the same 20x allowance as Pro, it costs $25 per unit of Plus-equivalent usage, against $10 on the $200 plan. To match the $200 plan's rate it would need 50x Plus. You would be paying for latency, not quota.
- Cerebras is an inference, not a leak. Nothing in the strings mentions it. The link rests on OpenAI's real $10B+ Cerebras deal and on GPT-5.6 Sol Ultrafast, which already runs on Cerebras at up to 14x standard speed.
- New buyers cannot get the $200 plan right now. OpenAI paused new Pro 20x sign-ups on September 10 after GPT-6 Astra demand. For anyone not already on it, the real choice today is $100 Pro or, if it ships, $500 Pro Max.
- The comparison tweets are out of date: Claude Max 20x is $200, not $250, and Google cut AI Ultra from $250 to $200 at I/O 2026. At $500 Pro Max would be the most expensive individual AI subscription from a major lab by 2.5x.
- The earliest plausible announcement is OpenAI DevDay on September 29. That is timing, not evidence.
Written September 25, 2026, before any announcement. Every claim below is labelled by what backs it: a first-party OpenAI or vendor statement, a leaked artifact, or a post on X with nothing behind it. The primary leak write-up is TestingCatalog; the pause is on the record in Fortune and TechCrunch. We will update this page when OpenAI says anything.
What leaked, and where it came from
On September 24, researcher Tibor Blaho posted that OpenAI is preparing a new, higher ChatGPT Pro plan. Within hours TestingCatalog, which tracks unreleased ChatGPT builds, reported the price: $500 a month, under the name ChatGPT Pro Max. Both point to the same source, subscription references shipped in ChatGPT's front-end configuration but not yet shown to users.
That matters for how much weight the leak can carry. A string in production code is a real artifact: somebody at OpenAI wrote it, and it survived into a build. It is also not a launch. Plans get written, priced, and shelved. The name, the price and one line of description are what the artifact shows. Everything else circulating on X was added on top.
The one line is "Fastest Work and Codex". Work is ChatGPT's mode for long, multi-step tasks, and Codex is OpenAI's coding agent, which can run on a single job for hours. Compared with the existing Pro plan description, TestingCatalog says that line is the only change.
The leak scorecard: confirmed, leaked, or tweet-only
Last week we graded the Claude Opus 5.5 leak on evidence strength and then let a weak evidence grade slide into a verdict that it was false. It was right on eight of nine claims (the scorecard is here). So this table grades two things separately: what backs each claim, and how likely it is to hold.
How much is ChatGPT Pro Max, and how it compares
At $500 a month, or $6,000 a year before tax, ChatGPT Pro Max would sit at the top of the consumer AI market by a wide margin. Several comparisons doing the rounds use stale numbers, so here is the current lineup, checked against each vendor's own pricing page on September 25.
Sources: Claude pricing, Google AI subscriptions after I/O 2026, and OpenAI's own ChatGPT Pro tier documentation. Coverage that puts Claude Max at $250 or Google AI Ultra at "roughly $250" is out of date. The three big labs have converged on the same ladder, $20, $100 and $200 for 1x, 5x and 20x. Pro Max is the first rung above it.
The math: $500 buys speed, not quota
The cleanest way to read a subscription ladder is price per unit of usage. Take the entry plan's allowance as one unit and divide.
Every ChatGPT tier until now got cheaper per unit as you climbed, or held flat. Pro Max breaks the pattern unless OpenAI quietly raises its allowance to about 50x Plus, and nothing in the leak says it will. If the allowance stays at 20x, you pay $300 more a month for the same amount of work, done faster.
That is not automatically a bad deal. It is a different product. The question stops being "how much can I run" and becomes "what is an hour of my waiting worth". At a loaded cost of $100 an hour for an engineer, the $300 premium pays for itself if faster Codex and Work runs save three hours of idle time a month. For someone who kicks off long agent jobs and waits on them, that bar is low. For someone whose bottleneck is quota, it is money for nothing.
Where the speed would come from
OpenAI signed a deal worth more than $10 billion with Cerebras in January 2026 for up to 750 megawatts of inference capacity through 2028, pitched explicitly as faster responses for reasoning models. In August the two companies previewed GPT-5.6 Sol Ultrafast, which runs the same Sol weights on Cerebras hardware at up to 14x standard speed and up to 750 output tokens per second. It launched as an API service tier for select customers, not as a ChatGPT feature.
A "Fastest" ChatGPT plan is the obvious consumer wrapper for that capacity, which is why the Cerebras link is plausible. It is still inference. The leaked strings do not name a vendor or a model, and there is no public sign yet of GPT-6 Astra, OpenAI's current flagship, running on Cerebras.
Why now: GPT-6 Astra broke the $200 plan
The context is a capacity crunch that OpenAI has been open about. GPT-6 Astra launched on September 3, and demand outran compute within days. On September 10 OpenAI paused new sign-ups and upgrades to the $200 Pro plan. Thibault Sottiaux, who leads core products, said the $200 subscriptions "put the most strain on our systems" and that OpenAI wanted "to take the smallest step that allows us to continue giving the broadest access possible" (Fortune). Existing subscribers kept their plans; the $100 Pro plan, API and business plans were not affected.
Around the pause, OpenAI reset usage for paid users on September 7 and then, according to multiple reports, tightened Astra allowances for the heaviest users a couple of days later. Astra also burns through a shared Work and Codex allowance faster than GPT-5.6 Sol does. On September 23 OpenAI opened a narrow door: people who held the $200 plan on September 10 can come back once within 30 days of losing it. New customers still cannot buy it.
Read together, the sequence explains the leak better than any conspiracy theory. OpenAI has more demand than capacity, its top plan is closed, and it has a fast-inference partner coming online. A higher-priced plan that sells priority and speed rather than raw volume is how you monetise scarce fast capacity without handing out more tokens you do not have.
The criticism, and what is fair in it
The loudest reaction on X is that OpenAI is squeezing lower tiers to make $500 look necessary. One widely shared summary recast the ladder as "$20 is the new free tier, $200 is the new $100". The fair part: limits on every paid tier are opaque, they change without notice, and OpenAI does not publish a numeric weekly ceiling for any plan. When the allowance is invisible, a price increase and a limit cut feel identical to the customer. The part that is not supported: there is no evidence the Astra limit cuts were made to sell Pro Max. A compute shortage that forced a sign-up pause is the simpler explanation, and it is on the record.
The spend shift behind the hype
The other number in the thread comes from Guillermo Rauch, CEO of Vercel, who posted that on Vercel's AI Gateway over the last two months Anthropic's share of spend fell from 69% to 40%, OpenAI's rose from 10% to 24%, and Kimi K3 plus DeepSeek took roughly half of what Anthropic lost. He credited GPT-6 Astra and GPT-5.6 Sol, and said OpenAI now leads on token count.
Treat it as a first-hand signal, not an index. It comes from the person who runs the gateway, which is why it is worth quoting, but it is a post, not a published dataset, and one gateway's customers are not the market. Vercel's own monthly production index for August put Anthropic at a higher share of spend, so the window matters. The direction is consistent with other reports that OpenAI briefly passed Anthropic in spend on OpenRouter in September. For the model-level picture see our running comparison of the frontier models, and for why Kimi K3 is taking share at all, our Kimi K3 breakdown.
What ChatGPT Pro Max means for developers and teams
For individual developers the decision is simpler than the noise suggests. If you run out of quota, Pro Max as leaked does not solve that problem, and the $100 Pro plan or a second vendor does it for less. If you sit and watch Codex work on long jobs, speed is worth real money, and Pro Max may be the first plan priced for exactly that.
For teams the more useful lesson is about dependency. In three weeks one vendor launched a flagship, reset limits, cut them, closed its top plan to new buyers, and prepared a plan at 2.5x the price. None of that was malicious and all of it changed what a seat costs and what it can do. If your engineering workflow runs through one consumer subscription, you inherit every one of those decisions. That is the argument for being able to move between Codex, Claude Code and Cursor without rewriting your process, and for knowing how to run an OpenAI model inside Claude Code when the pricing on one side moves.
It is also a reminder that agent speed and agent safety scale together. The agent-driven breach of OpenAI disclosed this month is a fresh example of how far a fast, autonomous agent gets before a human looks. A plan that makes Codex faster makes its mistakes faster too.
Stop paying for AI seats you cannot measure
Seat prices now run from $20 to a leaked $500, the allowances behind them are unpublished, and they changed three times this month. The only number that means anything is cost per finished task on your own work.
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 is an unreleased ChatGPT subscription found in OpenAI's front-end code on September 24, 2026. It is priced at $500 a month and its only listed difference from ChatGPT Pro is "Fastest Work and Codex". OpenAI has not announced it.
How much will ChatGPT Pro Max cost?
$500 per month in the US, or $6,000 a year before tax, according to the leaked strings. Secondary coverage reports a UK price of about £445 including VAT, which we could not confirm from a primary source.
When will ChatGPT Pro Max launch?
OpenAI has not given a date. The earliest plausible venue is OpenAI DevDay on September 29, 2026, but that is based on timing only. Plans found in code are sometimes changed or never shipped.
Does ChatGPT Pro Max have higher usage limits than Pro?
Not according to anything published so far. The leaked description only adds "Fastest Work and Codex". Higher limits are possible but unconfirmed. If it keeps the 20x allowance of the $200 plan, it costs $25 per unit of Plus-equivalent usage instead of $10.
Is ChatGPT Pro Max powered by Cerebras?
Unconfirmed. The leaked strings do not mention Cerebras. The link is an inference from OpenAI's $10 billion plus compute deal with Cerebras and from GPT-5.6 Sol Ultrafast, which already runs on Cerebras at up to 14x standard speed in the API.
Can I still buy the $200 ChatGPT Pro plan?
Not as a new customer. OpenAI paused new sign-ups and upgrades to the $200 Pro plan on September 10, 2026 because of GPT-6 Astra demand. Existing subscribers keep it, and people who held it on September 10 can return once within 30 days. The $100 Pro plan is still available.
How does ChatGPT Pro Max compare to Claude Max?
Claude Max tops out at $200 a month for 20x Pro usage, and it is open to new buyers. ChatGPT Pro Max would cost $500 with no published usage figure. At the leaked price it would be 2.5x the most expensive Claude plan.
Is ChatGPT Pro Max worth $500?
Only if speed is your bottleneck. If you regularly run out of quota, the leaked plan does not address that. If you spend hours waiting on long Codex or Work runs, the $300 premium over the $200 plan pays back once it saves about three hours a month at a $100 hourly cost.