OpenAI has just launched the Pro 500 plan priced at $500 per month at DevDay, while reopening the $200 Pro plan to new subscribers with much lower usage limits. A few weeks prior, sign-ups for the $200 Pro plan itself were temporarily paused due to Astra demand exceeding infrastructure capacity.
What’s in the new plans?
The three Pro plans are now divided into clear tiers:
| Plan | Price per month | Astra Ultrafast |
|---|---|---|
| Pro 100 | $100 | No |
| Pro 200 | $200 | No |
| Pro 500 | $500 | Yes |
Pro 500 offers the highest usage allowance among the three tiers and is the only Pro plan with access to Ultrafast, an Astra variant that generates tokens up to 8 times faster in Codex. OpenAI documentation notes a ceiling of around 300 tokens per second. Users select Astra and then choose Ultrafast in the model picker.
OpenAI also explicitly notes that Ultrafast consumes usage 8 times faster than standard speed, and after running out of usage, additional credit top-ups are charged at 6 times the standard rate. Thus, a more expensive plan does not proportionally translate to more work done. Purchasing top-up credits on Pro 100 or Pro 200 also does not unlock Ultrafast.
Pro 200 reopens with half the usage
This is much more frustrating than the new tier. Thibault Sottiaux, the engineer leading Codex, explained prior to the event that the usage calculation had changed; when translated to API cost, the value of the Pro 200 plan is now half of what it was before.
Although the price remains at $200, terms for new users have been significantly tightened:
- Overall limit: Reduced from 20 times down to 10 times compared to the ChatGPT Plus plan.
- GPT-6 Pro limit: The maximum number of ChatGPT messages dropped from 200 to 100 messages per week.
OpenAI justified this decision based on models achieving higher performance along with API price cuts for Sol and Luna. However, this argument only holds if your workload relies entirely on the newly discounted models. Conversely, users who depend on other models without API price reductions will be at a major disadvantage.
In addition, OpenAI provided specific rules regarding its grandfathering policy: Active Pro 200 subscribers at the cutoff date (or those maintaining a subscription within the previous 7 days) will keep their old usage limits until October 29, 2026, provided they maintain a continuous subscription. After this milestone, accounts will automatically transition to the lower usage limits, even though the cost remains $200/month. Those who do not meet the criteria will not receive further explanatory notices.
Two important points should be noted: Keeping the old limit does not mean being upgraded to Pro 500 or unlocking the Ultrafast feature. Furthermore, if a user had canceled the plan but reactivated it within 7 days prior to the cutoff date, they will still be eligible for the grandfathered usage limits through October 29.
Why these figures are easily misunderstood
A notable point is that official OpenAI channels have never released specific multiplier figures for Pro 500, leading many external articles to speculate. Although confirmed to be the plan with the largest usage capacity, the figure of “25 times Plus” appearing on comparison tables from some sources is actually just a calculation derived from price ($500 versus $20) rather than an officially published specification. Without the actual usage multiplier, this pure calculation does not accurately reflect actual usage value.
Therefore, the claim that “usage is 25 times that of the Plus plan” is merely a third-party perspective. The most accurate statement is: the Pro 500 plan costs 25 times as much, comes with superior processing speed advantages that the lower two tiers lack, while detailed capacity multipliers remain undisclosed.
For individual customers currently subscribing to the $200 tier, the core question is not whether to upgrade to $500, but rather the ability to manage quota consumption. OpenAI specifies that capacity is calculated based on a combination of factors: the model used, prompt complexity, context length, reasoning level, and accompanying tools; meanwhile, Codex and ChatGPT Work share a single quota allocation. This means an intensive coding session could consume the entire allowance intended for other automated tasks. Thus, without transparent usage figures for Pro 500, spending $500/month remains an uncertainty regarding the actual workload received in return.