OpenAI to Reopen $200 Pro Subscription, But Usage Value Halved in API-Equivalent Terms

Tibo (Thibault Sottiaux) announced that the $200 Pro subscription, paused for nearly three weeks, will reopen to new users tomorrow. What follows is not a price adjustment but a change in how usage is calculated: by his own account, the new plan converted to API spending amounts to only half of the old $200 Pro. In other words, the monthly fee is still $200, but the usage that money can buy—if purchased on the API at list prices out of pocket—is worth only half of what it used to be.

This is the starting point for understanding the entire controversy. Looking at this alone, users lose out, and they lose half, which is why it sparked an outcry. The core of the dispute lies in the phrase "API spending halved" being sufficiently vague: it sounds like saving users money, but in practice it roughly means cutting the token quota tied to the subscription in half. How much less work can actually be done still depends on what users buy "in the store."

If the new Pro quota is indeed calculated in API-equivalent dollars, then halving the budget while halving unit prices may not be as bad as it sounds on paper. GPT-6 Sol's API price dropped from the previous generation's $4 per million tokens input and $20 output to $2 input and $10 output; GPT-6 Luna dropped from $0.20 input and $1.20 output to $0.10 input and $0.50 output.

For users mainly working with Sol and Luna, the token volume they can run is roughly on par with a month ago, and combined with the new models' capability improvements, this may be what Tibo means by "can get more work done." It should be noted that this "roughly on par" calculation is based on the assumption that budget and unit prices halve in tandem—it is an inference from official statements and price data, not a promise given by officials.

One model is not in this equation: flagship GPT-6 Astra's API price remains $10 input and $50 output, with no reduction this time. For heavy users who subscribed to the $200 Pro precisely for Astra, by the same calculation, usable capacity is halved in real terms.

This constitutes the most clearly harmed group in this adjustment. Sol and Luna price cuts can offset the halved budget, but Astra not dropping means these users cannot hedge the quota shrinkage through lower unit prices.

Facing dissatisfaction, Tibo gave four reasons in his tweet explaining why users can do more work after getting the new Pro than a month ago. First, OpenAI will not discount in other ways, and he promised not to reintroduce the 5-hour limit; Plus users already regained the 5-hour window at the end of August, while Pro users can continue at their own pace, using up their weekly quota whenever they want.

Second, he promised that the workload and quality subscribers get will only rise over time: models become more efficient, OpenAI will pass the dividends through in the form of API price cuts, and the value of each dollar for subscribers will rise accordingly. Third, OpenAI doesn't want to create an incentive for itself to "raise API list prices and then circumvent them with discounts" to make subscriptions look like a good deal; instead it wants to pursue continuous price cuts and continuous capability improvements. He cited GPT-6 Sol and GPT-6 Luna, both launched at half the price of their predecessors. He even predicted that prices will eventually fall low enough that most people buying usage on demand will suffice, and there won't be much difference in "how much you get per dollar" between subscriptions and the API.

Fourth, tomorrow the subscription will add some new things that don't consume usage—as for what, he kept it a mystery. All four points are Tibo's public statements, official promises rather than accomplished facts.

Why OpenAI made this cut is not hard to understand from the timeline. After GPT-6 Astra launched on September 3, demand surged; on September 10, OpenAI paused new subscriptions and upgrades for the $200 Pro. Tibo's explanation at the time was that this tier puts the most pressure on the system.

The scale of the subsidy had been estimated externally before. SemiAnalysis estimated that a maxed-out $200 Pro subscription could equate to about $14,000 per month in usage at API prices—the subscription's subsidy to heavy users far exceeds outside imagination. After Astra pushed compute to the point where subscriptions had to be paused, tightening this subsidy was almost inevitable. It should be noted that the $14,000 figure comes from SemiAnalysis's estimate, not data published by OpenAI.

An earlier signal appeared on the pricing page: the "5x" and "20x" in the Pro tiers' usage descriptions were deleted, leaving only a vague "More usage than Plus," and a suspected $500 Pro Max appeared in a code repository. In hindsight, erasing the explicit multipliers may have been making room for today's adjustment.

The first unresolved question: does the new calculation method apply only to new subscribers, or will it also cover existing $200 Pro users? There is currently no conclusion. This directly determines whether the adjustment affects only incremental users or all existing $200 Pro users.

The second is what exactly the "new things that don't consume usage" added tomorrow are—Tibo did not disclose. The third is whether Tibo's claim of "can get more work done" holds up in actual use, which remains to be verified. In addition, whether the suspected $500 Pro Max tier actually exists and when it will launch also have no answers.

Tibo said at the end of his tweet that he wanted to make things clear before tomorrow's series of big launches—before drowning everyone in good news, let everyone first understand this change. Tomorrow's OpenAI Developers Day will bring a "rain of good news"; whether it can smooth over this cut will soon be clear.