Skip to content

Possible Codex usage/reset bug — unexpected drop from 98–99% to 16% after a reset #44206

Open

Description

@lyh0825

What version of the Codex App are you using (From “About Codex” dialog)?

26.903.61454

What subscription do you have?

PRO x20

What platform is your computer?

Darwin 25.6.0 arm64 arm

What issue are you seeing?

Subject: Possible Codex usage/reset bug — unexpected drop from 98–99% to 16% after a reset

Hi Codex Support Team,

I’m a Pro subscriber using the Codex desktop app. I’d like to report a possible usage-reset or usage-reporting bug and request that my reset credit be restored, along with appropriate compensation for the disruption.

This happened approximately within the last one to two hours. I don’t have the exact timestamps, but the sequence was:

  1. While Codex was actively working on a task, it reported that my usage limit had been reached and stopped working.
  2. I clicked Reset at the bottom of the conversation, using one of my usage-reset credits.
  3. After the reset, the app still showed 0% remaining.
  4. I restarted Codex, and the remaining usage then showed 100%.
  5. After a brief conversation and one work session, it showed approximately 98% or 99% remaining. I don’t recall the exact percentage.
  6. Shortly afterward, the remaining usage unexpectedly dropped to 16%. This seemed inconsistent with the limited work I had just performed.
  7. I restarted Codex again, but the usage stayed at 16%.

The attached screenshot shows the 16% remaining state. It documents the current display; I don’t have screenshots of every earlier step.

There appear to be two issues: the reset initially failed to update the displayed usage until I restarted the app, and the newly restored allowance subsequently dropped sharply without an explanation. I cannot determine whether this was a display synchronization problem, a usage-accounting issue, or something else.

This interrupted my ongoing work, cost me productive time, and required additional restarts and troubleshooting. It also appears to have wasted one of my reset credits, since the restored allowance did not remain available as I expected.

I’m requesting that you:

  • Investigate the reset and usage records to explain what happened, including whether the reset was applied correctly and what caused the sudden drop.
  • Correct any inaccurate usage balance and restore the reset credit used during this incident.
  • Provide appropriate additional compensation, such as usage allowance or a service credit, for the interruption and lost productive time.

If you need more information, please tell me exactly what would help—for example, my app version, operating system, affected conversation or task IDs, approximate timestamps, or diagnostic logs—and how to provide it securely.

Thank you. I’d appreciate a clear explanation and a resolution addressing both the usage discrepancy and the reset credit.

What steps can reproduce the bug?

Feedback ID: 01a08710-779c-7b53-abbd-e25d544bb640

What is the expected behavior?

No response

Additional information

my codex account
aivorx
@aivorx

login method gmail
drinkle1226@gmail.com

Activity

  1. added
    appIssues related to the Codex desktop app
    on Sep 9, 2026
  2. added
    bugSomething isn't working
    rate-limitsIssues related to rate limits, quotas, and token usage reporting
    on Sep 9, 2026
  3. github-actions commented on Sep 9, 2026

    @github-actions
    Contributor
  4. dmognaschi commented on Sep 9, 2026

    @dmognaschi

    The same to me: from 30% to 0% in a single prompt a few minutes ago.

  5. gmercey commented on Sep 9, 2026

    @gmercey

    same here: from 70% to 20% while using astra light with one simple prompt and quick answer (not fast mode)

  6. Belgian-Coder commented on Sep 9, 2026

    @Belgian-Coder

    Same for me went from 99% to 2% after updating the desktop app. looks like it lost the last reset.

  7. exitoenlafamilia commented on Sep 9, 2026

    @exitoenlafamilia

    Same here! After a reset got 13% reduction after a few minutes. Im pro 20x plan

  8. capinfl commented on Sep 9, 2026

    @capinfl

    I feels like everyone got their usage vanished. I had GPT6 astra for limited tasks, havign 5.6-sol for the main tasks, even tho I did a weekly reset today it is simply gone after a few hours of usage.

  9. bgyss commented on Sep 9, 2026

    @bgyss

    This happened overnight for me on a Plus membership - I only run 2 everyday automations and my weekly went from ~80% to 7% between 11pm and 10am this morning.

    Here's my feedback ID in case anyone at OAI is interested in my logs: no-active-thread-01a0872c-a7fd-7201-841e-d192ab67f907

  10. xXAvoraXx commented on Sep 9, 2026

    @xXAvoraXx

    +++

    I had the same problem; my Codex limits were full, so I used my reset option. At first, the limit showed as 0%, then it went to 100%. After using it for a bit, it dropped to around 98%, and then immediately went back to 0%.

    As a result, I used my reset option, but it was automatically deleted before I could use the entire limit, and it dropped to 0%.

  11. tradesdontlie commented on Sep 9, 2026

    @tradesdontlie

    Additional Windows evidence for the September 9 sudden weekly quota loss. User reports roughly 65% -> 8% remaining with no agents running. Local recorded rate-limit payloads substantiate 66% -> 8%, although they do not capture the exact instant of the change.

    Environment: Windows x64; ChatGPT/Codex desktop host; bundled codex-cli 0.153.4 (queried directly); plan_type=pro; recent session model gpt-6-astra. Desktop UI version not independently verified. No subagents launched for this investigation.

    Exact quota evidence (UTC)

    Field Last prior general Codex snapshot First investigation snapshot
    Recorded at 2026-09-09T02:14:45.638Z 2026-09-09T17:25:03.470Z
    limit_id codex codex
    primary.used_percent 34 92
    Remaining 66% 8%
    primary.window_minutes 10080 10080
    primary.resets_at 1789435451 1789409560
    Reset in UTC 2026-09-15T01:24:11+00:00 2026-09-14T18:12:40+00:00
    secondary null null
    plan_type pro pro
    credits has_credits=false, unlimited=false, balance=0 same

    The weekly reset moved EARLIER by 25,891 seconds (7h 11m 31s), alongside a 58 percentage-point increase in used quota. This is the same general Codex weekly bucket, not a comparison against the separate Spark bucket or a five-hour window. Adjacent prior snapshots varied by one second in reset time; that does not explain this seven-hour change.

    Local activity checks

    • Scanned locally present session and archived-session JSONL files modified since September 8 for task lifecycle events and token_count.rate_limits payloads.
    • Last recorded preceding task completed September 9 at 02:26:13.660Z (September 8, 22:26 EDT).
    • No task lifecycle activity was found in those records between that completion and this investigation starting at 17:25:01.236Z (13:25 EDT).
    • The 92% used snapshot was recorded at 17:25:03.470Z, at the beginning of the investigation, before its substantive diagnostic work.
    • Desktop host and two Codex processes started around 17:21-17:22Z. Process presence is not proof of active model requests.
    • Earlier sessions did contain substantial legitimate usage, including a long session with cumulative input tokens heavily cached. Cumulative session totals are not equivalent to new usage during the idle interval, nor can they be converted into allowance percentages without the backend ledger.

    Interpretation and limits

    This corroborates a quota/accounting/window synchronization discrepancy and is consistent with the other reports today. It does NOT establish a specific backend bug. Possible explanations needing investigation include reset/window reassignment, reset rollback, delayed usage reconciliation, or an allowance denominator change. No claim is made that these are confirmed. Local logs cannot rule out usage on another device/cloud, account/workspace changes, or missing telemetry. No reset action or plan change was independently established in this investigation. We cannot reproduce on demand.

    Requested investigation

    Please correlate the timestamps with the account-side usage ledger and explain:

    1. Why did the weekly reset epoch move backwards by over seven hours?
    2. Which requests/adjustments caused 34% -> 92% used, and when were they actually incurred?
    3. Did a rollout, reset rollback, allowance change, or delayed reconciliation affect users around 16:00-17:25 UTC September 9?
    4. If the ledger is incorrect, please restore the affected allowance without consuming a user reset.

    This report intentionally excludes authentication data, account identifiers, private prompts, repository content, and raw logs. Detailed local evidence is preserved and can be supplied privately to OpenAI Support if needed.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    appIssues related to the Codex desktop appbugSomething isn't workingrate-limitsIssues related to rate limits, quotas, and token usage reporting

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions