Thibault Sottiaux, head of OpenAI Codex, stated that GPT-6 Astra has been released to users ahead of schedule and will provide a full quota reset for all Plus, Pro, and Business users before the end of the day; accounts that are newly created or upgraded before 8 PM Pacific Time will also be eligible.

The quota reset is not a permanent increase but resets the usage limit stored in the account, which users must manually redeem to refresh the five-hour window and weekly quota. The weekly reset day will be approximately delayed by seven days, with a common validity period of about 30 days. The help center specifies that on September 3 and 4, existing eligible paid users will each receive one reset; the deadlines for new accounts are 10 PM and 8 PM respectively.

Previously, due to the staggered launch of Astra, Pro users complained that high-priced seats did not receive priority access to the model. Sam Altman publicly apologized and described the rollout as "chaotic." Tibo had previously promised that if paid plans did not receive Astra daily, a quota reset would be provided. The team also had to urgently fix code due to automatic reviews and backend suggestions that repeatedly consumed tokens and issued resets.

Active Codex users surpassed approximately 20 million and 25 million in August, with the official team frequently using resets as compensation and celebration tools. Astra's enterprise workspaces are defaulted to closed, and Plus and some Business seats receive funds more slowly, making resets a buffer inventory.

In market mechanics, the buyers are programming users who have paid a monthly fee but are stuck at the quota wall, while the sellers are model companies that exchange one-time resets for retention. This is a quota subsidy driven by delivery delays: funds are not returned to bank cards but remain in accounts as unfulfilled reasoning minutes. The beneficiaries are heavy users who can immediately redeem resets to run Astra or older models, while the pressured parties are the engineering teams responsible for computing resource scheduling and must switch on the models before the weekend.

Source: Public Information