[Usage limits] GPT-6 Astra Low/Standard: 5h allowance exhausted after ~20–30 min of resumed use (Oct 8) #51951

Summary

A ChatGPT Plus user reports a sudden and unusually rapid exhaustion of their Codex 5-hour allowance after only approximately 20-something minutes of resumed activity on Oct 8, 2026. This is an unverified usage-accounting concern, not proof of incorrect billing or reduced allowance. Please investigate server-side usage reconciliation.

Environment (confirmed from screenshot)

• Client: Codex inside ChatGPT app on a tablet, not Desktop or CLI
• Subscription: ChatGPT Plus
• Model: GPT-6 Astra
• Reasoning effort: Low
• Speed: Normal / Standard (NOT Fast)
• Timezone: Asia/Shanghai (UTC+08:00)

User-reported timeline

• User began using Codex at approximately 01:00 on Oct 8 (UTC+08:00); by around noon the weekly quota displayed 70-something percent remaining, roughly 20–30 percentage points depleted.
• User resumed activity at approximately 11:00 and now clarifies that after only ~20-something minutes of activity (under 30 minutes), the 5-hour allowance displayed exhausted. Earlier wording saying "under an hour" understated the reported rapidity.
• During the previous allowance cycle, the user used the same Astra model and recalls significantly slower consumption with a comparable development workflow.
• The screenshot of the current settings shows Astra / Low / Standard. The user does not want to switch models.

Important limitations: We do not have the account's exact 5-hour window reset/start time, before/after allowance snapshots for this 20–30 minute period, per-request logs, or authoritative server ledger. Thus the 5-hour allowance displayed exhausted after 20-something minutes of resumed activity, but we cannot establish that a full fresh 5-hour bucket was consumed exclusively during those minutes. Do not assume the Oct 5 speed update caused overcharging.

Requested investigation

1. Reconcile backend per-turn usage, actual remaining quota before/after the reported 11:00–11:30 interval, and 5-hour/week allowance window start/reset times for Oct 7–8 (Asia/Shanghai).
2. Clarify whether the recent Astra default generation-speed increase impacted quota costing/weighting versus only throughput.
3. Inspect background/parallel work, retries, hidden task execution, context replay, caching changes and any rate-limit/accounting discrepancy, including whether Normal speed in mobile UI matched the served tier.
4. Provide an explanation of the charges and restore incorrectly deducted allowance if a metering error is confirmed.

No account identifier, keys, private repo names or sensitive screenshots were attached publicly. The user can supply account-specific evidence through secure Support.

Related reports/tracker: #41220, #51604.