This weekend, Coding Agent users saw two very different "quota adjustments" at the same time.
On the one hand, the head of the Codex, Tibo (Thibault Sottiaux), presses again the "reset button" that is already being played by users: all the users who pay for the Codex and ChatGPT Work will receive a reset.
According to Tibo, the team recently reviewed thousands of user reports and focused on fixing a set of "stealthy-eat-quota" issues. Depending on the user's specific usage, the same Codex quota is expected to run 10% to 50% longer than before.
On the other hand, Anthropic also announced Claude Code's new quota scheme on the same day: it looks like a permanent addition, but it's actually a "degradation".
What's going on?
Claude Code：
It's a permanent increase of 25 percent, but now it's actually going to be a "degradation".
Anthropic announced that from September 14, Claude Code's "Standard Weekly Allowance" will be permanently increased by 25%, covering Pro, Max, Team, and the seat-based Enterprise package.
From a historical standpoint, it sounds like 25% off.
But the problem is that Claude Code users now enjoy a temporary +50% weekly allowance.
Anthropic itself calculates the bill very directly: Compared to today, Claude Code's weekly allowance will actually be reduced by 17%.
Therefore, a more accurate understanding of this adjustment should be that Claude Code ends the temporary expansion of its current size while permanently retaining part of its quota.
Anthropic's explanation is also realistic. The team thanked users for supporting the tests over the past few years and said they have been exploring "how much service can be delivered sustainably in the future".
At the same time, Anthropic predicts that it will launch a new usage mechanism that will allow users to "feel more when using Claude" while being able to see and control their consumption more clearly.
Almost at the same time, the two most-watched players in the field of Coding Agent began publicly discussing a now increasingly intractable issue: quantity.
When the agent changes from a few lines of code occasionally to hours of continuous work, calls dozens of tools, pulls up multiple sub-agent productivity tools, how smart the model is is the first question.
The following question has been raised before all users: where did my quota go?
Anthropic's statements are so clever that users may even think they're "25 percent less" when they look at it; another user jokes, "Wow, it's good we only lost 17 percent".
Users also immediately began to consider using OpenAI and Google's models, saying, "You're so folded up now that the next 5.1 series is better off, otherwise I have no reason to subscribe".
Compare Anthropic and OpenAI: The former will reduce the current Claude Code week quota by about 17%, the latter will redeploy users for Codex and ChatGPT Work, while fixing several "stealing quota" bugs so that the same allowance can be increased by up to 50%.
