Just now, the god of reincarnation, Tibo, has launched again, but this time it is not reincarnation.
Tibo (Thibault Sottiaux) announces that the $200 Pro subscription, which has been suspended for nearly three weeks, will be reopened to new users tomorrow.
The result was an adjustment in the way the consumption was calculated. According to him, if you take it seriously, the new program is calculated as an API expense, equivalent to only half of the $200 Pro of the old version.
The Chinese-English version of Tibo's tweets
Well, what exactly does that mean?
The key is these words: "Counting in API dollars".
In other words, the monthly fee is $200, but the amount that can be used to replace it is only half the value of the previous one if you get paid to buy it on the API at a nominal price.
For example, if the old program spent $2,000 on the API for a month, the new program spent only $1,000.
It's understandable to imagine a subscription as a shopping card. Formerly, it cost $200 to work on a card, and Carrie could browse the $2,000 allowance; from tomorrow, it'll cost $200 as well, with only $1,000 left.
It's just that users are losing, and they're losing half of it.
After all, the phrase "halving API spending" is vague and sounds like saving money for users, but it's actually about halving the amount of token subscriptions.
Of course, as to how little you can actually do, look at what you buy in the "store".
More specifically, GPT-6 Sol's API price has dropped from $4 per million of its predecessor's tokens for input, output, and $20 to $2 and $10, while Luna's has dropped from $0.20 and $1.20 to $0.10 and $0.50. If the new Pro's allowance is actually calculated in the same dollar as the API, the budget will be halved, and the single price will also be halved, with the number of tokens that can run roughly equal to a month ago, and the ability to superimpose the new model, which is perhaps what Tibo says: "I can do more work".
Tibo also gave four reasons for this in a tweet explaining why people will be more active after getting the new Pro than they were a month ago.
First, OpenAI will not give discounts in any other way, and pledges to no longer introduce a 5-hour limit. Note that Plus users have re-used the 5-hour window at the end of August, and Pro users can continue at their own pace, thinking about when to use the full-week quota.
Second, he promised that the amount of work and quality subscribers get will only increase over time. The model is becoming more efficient, OpenAI will pass on dividends in the form of API discounts, and the value of each subscription user's dollar will follow suit.
Third, OpenAI doesn't want to give itself the incentive to "raise API prices, come back with discounts" to make subscriptions look cheap. They want to do more than just keep the price down, keep the capacity up. Tibo's examples are the GPT-6 Sol and GPT-6 Luna, both of which were half the price they were before. He even predicts that prices will eventually drop to enough for most people to buy on demand, and that there won't be too much of a gap between subscriptions and APIs "how much a dollar can buy".
Fourth, tomorrow's subscription will include something new that doesn't consume. So what, he sold a knot.
But one model is not in this equation: the API price for the flagship GPT-6 Astra is still $10 for input and $50 for output, and this time it hasn't dropped.
SemiAnalysis has previously estimated that a $200 Pro subscription that is used up to a limit, which can be calculated at about $140,000 per month based on the API price, is a subsidy for heavy users that is far beyond anyone's imagination.
Prior to this, on September 10, OpenAI suspended new subscriptions and upgrades to the $200 Pro following the September 3 release of the GPT-6 Astra, which Tibo explained at the time was the biggest pressure users put on the system.
Two days ago, we reported a "silent operation" on the OpenAI pricing page: The "5x" and "20x" in the Pro's two dosage descriptions were removed, leaving only the vague "More usage than Plus" and a code repository with a suspected $500 Pro Max.
Looking back, erasing the clear multiples, maybe that's what's adjusting the space for today.
There is also an unresolved question: will the new calculation only apply to new subscribers, or will it simultaneously cover existing $200 Pro users?
At the end of the tweet, Tibo said he wanted to make it clear before tomorrow's big release, before flooding everyone with good news, before everyone understood the change.
Tomorrow, OpenAI Developer's Day, the "rain of good news" will wipe out the sword, and it will soon be clear.
This tweet is also in the style of Tibo: "Codexingly".
Do you want to re-order the $200 Pro?