GPT-6 Astra users report a quota glitch draining weekly capacity

• A routine game-development session on a 20× plan reportedly fell from 55% to 8% remaining in 30 minutes at Astra Medium, according to AIandDesign.
• OpenAI publicly said it was investigating unexpected Codex usage-limit resets, as LLMJunky's status-page screenshot captured.
• One account's percentage was restored, AIandDesign's follow-up said, but that user later reported Medium appeared to consume capacity at Ultra's rate in a later post.
• An Astra terminal session also displayed server_is_overloaded errors in thekitze's terminal capture.

OpenAI's Astra usage guide says Work and Codex draw from a shared allowance with five-hour and weekly windows. Its banked-reset explainer defines those resets as one-time replenishments, not a permanent quota increase. A Developer Community report separately described a single light Astra agent taking an account from roughly 78% to 3% in under a minute.

AIandDesign said their remaining weekly capacity fell 47 percentage points while using Astra Medium on a 20× plan.

They described the work in a follow-up as the same game project they had been running for some time, with nothing unusual about the session.

Other accounts described markedly different usage:

• Kaigani wrote in a quota report that asking Astra to quality-check work could burn 80% of weekly capacity.
• AIandDesign estimated in a pricing reply that a 45% weekly draw was roughly $20 to $25 on their plan, then said API credits available after the limit ran out cost much faster.

LLMJunky shared an OpenAI status-page screenshot saying the company was investigating unexpected usage-limit resets affecting some Codex users.

AIandDesign said in a follow-up that the percentage had returned to its earlier level, then called the event a glitch in another reply. The status wording names the reset issue but does not specify its cause or scope.

The account-level concern persisted after that apparent correction. AIandDesign wrote in a later post that Astra Medium still seemed to drain capacity as quickly as Ultra.

OpenAI's usage documentation describes two separate constraints: a five-hour allowance that begins with a new Work or Codex message after the prior window ends, and a weekly allowance that caps included use across the period. Both must have capacity remaining.

A banked reset is a saved, one-time limit reset rather than purchased credit or a permanent increase. In a contemporaneous reply, AIandDesign separately cautioned that an anticipated reset was not a banked reset.

A terminal capture from thekitze showed two server_is_overloaded responses during an Astra High session.

Thekitze had already written in an earlier post that Astra was constantly at capacity even on a $200 plan. In a later joke, he imagined an agent burning seven billion tokens over five hours and still missing a center this div request.

Rowan Cheung gave Astra quota a zero-out-of-10 rating in an Alps post. The posts document both allowance and availability complaints, but none identifies a shared cause.

Longer autonomous runs produced a different range of reports. LLMJunky said an Astra Extra High Fast goal had run for more than 18 hours while staying under 45% of the weekly allowance.

Stevibe reported roughly 40% of a Pro weekly quota after about 20 hours at Astra Extra High in a usage report, and called that amount generous in a reply. Neither post includes a token count, so they do not yield a comparable per-task burn rate.