On September 29, OpenAI wrote to some of the subscribers of its $200 Pro plan. The price did not change. Their Work and Codex quotas were to be reduced from 20 to 10 times those of the Plus plan, and their weekly messages to GPT-6 Pro from 200 to 100. Existing subscribers kept the old regime until October 29 inclusive. The day before it went into effect, OpenAI canceled everything.
Thibault Sottiaux, who heads Codex and ChatGPT Work at OpenAI, announced on October 29 that the company was giving up reducing the Pro 200 after feedback from its customers, and that it was doubling the Work and Codex quotas for the Pro 100 and Pro 500 plans, according to The Decoder. The Plus package, at $20, is not affected. OpenAI warns that a doubled quota does not double the number of tasks completed, as consumption varies depending on the model, task and settings.
One month to announce a drop, one day to replace it with an increase: the episode would be anecdotal if it were isolated. He is not. At OpenAI as at Anthropic, the price of the packages has not changed since the spring. Everything else has changed, several times.
Four resets in seven weeks at OpenAI
The decline of October 29 comes at the end of a series. OpenAI reset the Work and Codex counters of its paying subscribers on August 8, then on August 29, then those of all paid subscriptions on September 7, before paying a “reserve” reset on September 22, which the subscriber triggers himself, according to the AI Catchup statement. On August 29, Thibault Sottiaux explained that bugs in the count were inflating consumption and that, once corrected, the same quota would go 10 to 50% further. In early October, it promised 28 days of daily improvements or complete resets.
In the meantime, on August 25, the Plus package had regained its five-hour limit. The Pro 200 reduction went further: it halved what a high-end package contained on paper. She didn’t last a month.
Anthropic experienced the same back and forth
On April 21, Anthropic’s pricing schedule removed Claude Code from the $20 Pro plan. Less than 24 hours later, after an outcry from developers, the box was restored, developer Pasquale Pillitteri documented. On May 6, Anthropic doubled Claude Code’s five-hour limits and removed the peak hour discount, relying on leasing the entire capacity of SpaceX’s Colossus 1 data center, more than 220,000 GPUs. Since September 1, Claude Fable is only included in Max packages up to 50% of the weekly quota. On October 7, Anthropic added monthly credits for its API platform to Max plans, $100 on Max 5x and $200 on Max 20x.
These quotas also have a legal cost. A class action filed on June 15 accuses Anthropic of delivering, on its Max plans, much less usage than the promised multiples, Engadget reported. Anthropic declined to comment to the site. His request to dismiss the complaint will be examined on November 6.
Why does the quota move and not the price?
The IPO prospectus made public by Anthropic on October 26 gives the key, on Anthropic’s side at least. Subscriptions brought in $3.2 billion in the first half of 2026, 20% of revenue, with the rest coming from consumption billed to the token, according to the S-1 filed with the SEC. The same document shows where the money is made: in the third quarter, the 5% of Claude Enterprise users most active on Claude Code spent approximately twenty times more than those who do not use it, who remain under $75 per month.
A developer who codes every day is therefore both the most expensive subscriber of a fixed plan and the best prescriber of company contracts. Discouraging him with a quota that is too tight means sending him to the competitor. Letting it consume without limit is burning a rare calculation: Anthropic recognized that its cloud providers had delivered less than it asked for, as ActuIA noted in the draft prospectus. Between the two, the displayed price cannot move quietly. The quota is not published anywhere in absolute value, and is adjusted silently.
Anthropic spells it out in its risk factors: it has modified, and may modify, its usage limits, its flow limits and the capacities of its products. OpenAI does not have a public prospectus. The Decoder links its action of October 29 to competition with Anthropic, as the latter’s IPO approaches.
What looks like excitement is arbitration. In May, a new data center pushed Anthropic’s limits. In September, OpenAI wanted to trim those of the Pro 200, before giving up under pressure from its customers. The displayed price has become the only stable data for these packages. It’s the one that provides the least information.
What a subscription quota involves
Nothing, or almost. A team that equipped its developers with Pro or Max packages purchased access whose volume can be halved by simple email, with thirty days’ notice, then maintained the day before the deadline. Uber discovered the problem from the other end in June, capping Claude Code and Cursor after exhausting its budget in four months.
The individual package remains the cheapest tool for testing and for moderate use. Beyond that, the volume is negotiated where it is written: a company contract or consumption at the API, with quantified limits, a duration and a notice period. On November 6, a California judge will say whether the complaint against the promise of a usage multiple can go further.
Our articles will be prioritized in Google’s “Featured” section.
