Limits did not match plans for about 40,000 accounts
Thibault Sottiaux, who leads OpenAI's programming assistant "Codex," said on X on Oct. 20 (early Oct. 21 Japan time) that there had been a bug in Codex usage limits. Over the weekend, a string of users had reported that their usage allowances were running down faster than expected. He explained that for about 40,000 accounts, the usage limits being applied did not match the plans they subscribe to.
The bug has already been fixed. Affected users were contacted individually and given usage credits as compensation. He was also reported to have acknowledged that the company mishandled the matter.
Affected plans and direction of the error not disclosed
The company has not disclosed which plans' users were affected, whether the limits were stricter or looser than intended, or the details of the technical cause. It is also unclear what share of all Codex users the roughly 40,000 accounts represent.
Repeated confusion over usage limits
Complaints about how quickly Codex usage allowances run down had surfaced before. In June, after reports that allowances were dropping even when no work was being done, OpenAI temporarily reset usage limits for all users. Posts about allowances suddenly dropping have continued on OpenAI's official forum.
Starting Oct. 5, Sottiaux had promised to deliver either a Codex improvement or a usage limit reset every day for 28 days. On Oct. 16, he explained that context is now reused among helper AIs so that more work can be done within the same limits. Starting in December, OpenAI will combine the usage limits for ChatGPT and Codex into a single allowance.
