OpenAI may be preparing a $500-a-month plan called "ChatGPT Pro Max." TestingCatalog reported on September 24 that the description of an unannounced plan included wording indicating the fastest access to Work, which hands tasks off to the AI, and Codex, its development assistant. New sign-ups for the current $200 Pro plan are paused, and now a price 2.5 times higher has appeared. However, the official terms have not been disclosed. Comparing the existing pricing structure with the high-speed processing specifications shows that judging whether $500 is worth it requires separating response speed from the amount you can keep using.
The $500 Plan Versus Today's Pro
The central change TestingCatalog reported is a label reading "Fastest Work and Codex." The $600 price shown in the video it posted includes value-added tax (VAT); the outlet says the monthly price before tax is $500. Pricing in Japan and the launch date are unknown.
This label does not reveal the full specifications of the new plan. It is also unconfirmed how far usage limits will rise, so it cannot be read as "pay $500 and use it without limits." The name and price are both pre-announcement information.
The baseline for comparison, the existing plans, can be checked in OpenAI's Pro help page and Codex pricing documentation.
| Plan | Monthly price (USD) | Usage relative to Plus | Status as of September 26 |
|---|---|---|---|
| Plus | $20 | Baseline | Available |
| Pro 5x | $100 | 5x | Open to new subscribers |
| Pro 20x | $200 | 20x | New sign-ups and upgrades temporarily paused; existing subscriptions continue |
| Pro Max | Reported at $500 | Unconfirmed | Display of an unannounced plan has been reported |
Prices are compared before tax and do not show the final amount billed in each region. Usage multiples are also not a guaranteed number of messages; consumption varies with the model and the type of work.
The current $100 and $200 Pro plans share core features, and the main difference is the usage allowance. By contrast, the description of Pro Max reportedly puts speed front and center. The monthly price rises 2.5-fold ($500 ÷ $200), but that alone does not mean processing volume or performance will also rise 2.5 times.
In Fast Mode, Usage Also Drains Differently
According to the official pricing documentation, Work and Codex share usage limits and credits. Working on separate screens does not stack up separate, independent allowances. Moreover, tasks with long histories or complex reasoning consume more of the allowance than short requests.
Speed settings also come at a cost. OpenAI's speed specification explains that using Fast, the high-speed mode, increases credit consumption. Credits here are a billing unit consumed according to the amount of AI processing. A high-speed setting also uses up the allowance included in a flat-rate plan faster than usual.
| High-speed category | Target | Officially stated speed | Consumption / availability versus standard processing |
|---|---|---|---|
| Fast | GPT-5.6 | 1.5x | Credit consumption rate is 2.5x |
| Fast | GPT-6 Astra / Sol / Luna | No multiplier listed on the specification page | Credit consumption rate is 2.5x in supported environments |
| Ultrafast | GPT-5.6 Sol | Up to 14x, up to 750 output tokens per second | Limited preview, first available via API; a separate category from Fast in personal plans |
| Pro Max speed-up | Unconfirmed | Unconfirmed | Reported display describing it as "fastest"; specific allowance and consumption terms unconfirmed |
The table lines up, by category, the speed specifications as of September 26, the August 13 Ultrafast announcement, and the TestingCatalog report. It is not a performance test under identical conditions, nor does it directly rank the speeds of different models.
The 5x and 20x of the current Pro plans are multiples of usage allowance, while the 2.5x of Fast mode is a multiple of the consumption rate. From Pro Max's $500 price, no multiple of usable volume or speed can be calculated.
For example, even when asking the same model to do a similar amount of processing, choosing a faster setting drains the allowance more quickly. Even with a larger allowance, depending on the consumption rate of high-speed processing, the amount of work you can get done will not necessarily grow at the same rate. Evaluating Pro Max requires not only the monthly price and the "fastest" label but also the conditions for how much you can keep using at that speed.
Cerebras Is a Plausible Backdrop, but Its Link to Pro Max Is Unconfirmed
OpenAI is already moving low-latency inference using Cerebras toward practical use. According to the announcement of the partnership between the two companies, Cerebras's systems concentrate compute, memory, and high bandwidth on a huge chip, reducing the bottlenecks that slow inference in conventional configurations. It is infrastructure meant to shorten the wait from when a model processes input to when it generates an answer.
A concrete example is the Ultrafast version of GPT-5.6 Sol that OpenAI unveiled on August 13. The company indicated speeds up to 14 times faster than standard processing, at up to 750 output tokens per second. Tokens are the units in which a model handles text and other data, and they do not match the number of Japanese characters. It launched as a limited preview for the API, and the company says it will broaden the audience as capacity expands.
TestingCatalog cites this existing effort as a reason Cerebras might also be used for Pro Max. However, no official announcement linking the two has been confirmed. The published Ultrafast figures cannot be treated as Pro Max's speed or as GPT-6's performance.
In development work, you repeat a cycle of generating code, reading test results, and fixing it. If the model responds faster, the waiting time at each step can be shortened. On the other hand, test run times, responses from external services, and the time for a person to review results do not shrink at the same rate. Faster output per second and the time until the whole job is finished are separate metrics.
This difference shapes when paying extra for speed makes sense. It is likely to pay off for work where your hands stop every time you wait for an answer. Conversely, if most of the time goes to waiting on long tests or external processing, speeding up only the model's output will limit the overall reduction.
What You Need to Know to Judge the Extra $300 a Month
If we assume it arrives at $500 as reported, the difference from the $200 Pro plan would be $300 a month. For existing users, the yardstick is how much more work that spending adds and how much waiting time it removes.
People who hit their daily limits first will need to look at Pro Max's usage allowance and the consumption terms of high-speed mode. Current Plus and Pro plans also offer the option of buying additional credits after reaching the limit, so the cost can be compared with staying on the current plan and adding what you need. Usage with an API key is billed separately at API rates and is not calculated the same way as flat-rate usage within a personal plan.
For those who want to cut time spent waiting for responses, the first question is whether the models they normally use are covered by the speed-up. After that, it will be worth checking completion time and consumption when giving it the same task. Even if a fast model is available, if the capability or usable volume needed for your use changes, the monthly price difference alone will not allow a comparison.
The treatment of existing contracts should also be considered separately. What OpenAI halted from September 10 is new sign-ups and upgrades to the $200 Pro plan; renewals of existing subscriptions continue. There is an exception allowing eligible users to return once, within 30 days, after their use ends through cancellation or similar. The Pro Max report does not mean current plans will be discontinued or that users will be forced to migrate.
What OpenAI needs to formally disclose are the models that will be sped up, the usage allowance, and the conditions under which that allowance is consumed. If those are in place, and you can keep using the volume you need while shortening waiting time in your everyday work, Pro Max could become an option for developers and researchers to buy back working time.