Skip to content

Possible overcounting of Codex weekly usage #44685

Open
Open
Possible overcounting of Codex weekly usage#44685
Labels
CLIIssues related to the Codex CLIbugSomething isn't workingrate-limitsIssues related to rate limits, quotas, and token usage reportingwindows-osIssues related to Codex on Windows systems

Description

@p-poppe
p-poppe
opened on Sep 11, 2026

Possible overcounting of Codex weekly usage

My weekly Codex allowance went from 0% to 100% between September 10, 2026, 17:07:59 CEST and September 11, 2026, approximately 02:00 CEST — less than nine hours.

Environment

  • ChatGPT self-serve Business workspace Premium PLAN
  • Codex CLI 0.154.0 on Windows
  • GPT-5.6 Sol
  • Standard service tier; Fast mode was not enabled
  • 11 user sessions and 8 subagent sessions

Local Codex telemetry

  • 306,015,436 input tokens
  • 300,207,488 cached input tokens
  • 733,962 output tokens
  • 2,499 token-count events
  • Server meter progression: 0% → 19% → 30% → 50% → 99% → 100%

At the published GPT-5.6 Sol rates, this is approximately 3,950 credits. Please audit the server-side usage ledger for this weekly window and verify:

  1. Whether cached input was charged at the cached-input rate.
  2. Whether concurrent sessions or subagents were double-counted.
  3. Whether repeated context during internal tool loops is intentionally charged for every model invocation.
  4. The exact request-level credit calculation that exhausted the weekly allowance.

My comparable workload in Claude Code using Sonnet normally exhausts its weekly allowance after approximately three days, including four or five concurrent sessions. Exhausting Codex in less than nine hours is therefore unexpectedly high.

I can provide sanitized aggregates and session IDs, but cannot share raw session transcripts because they contain confidential code and business data.

Activity

  1. github-actions
    added
    bugSomething isn't working
    CLIIssues related to the Codex CLI
    windows-osIssues related to Codex on Windows systems
    rate-limitsIssues related to rate limits, quotas, and token usage reporting
    on Sep 11, 2026

  2. github-actions commented on Sep 11, 2026

    @github-actions
    github-actionsbot
    on Sep 11, 2026 – with GitHub Actions
    Contributor

    Potential duplicates detected. Please review them and close your issue if it is a duplicate.

    Powered by Codex Action

  3. pvzv0134-dot commented on Sep 11, 2026

    @pvzv0134-dot
    pvzv0134-dot
    on Sep 11, 2026 – with ChatGPT Codex Connector

    Adding a similar ChatGPT Pro case with an independent Usage-page data point.

    • Plan: ChatGPT Pro
    • Date: September 9, 2026 (Europe/Berlin)
    • Model: GPT-6 in Codex; exact variant/reasoning level and client build were not captured
    • The ChatGPT Usage page reported approximately 290 million tokens for that day
    • Approximately 70 percentage points of the weekly Codex allowance were consumed during the day; a major portion of the depletion appeared to occur within roughly three hours
    • The remaining allowance was consumed later, and the weekly balance is now 0%
    • This was materially faster than the prior baseline on the same account, although I do not currently have a controlled historical comparison or request-level logs

    Important separation: there was also a separate temporary event on September 9 where the displayed balance went from roughly 40% remaining to 0%. That allowance was subsequently restored during the official usage-limit incident. I am not counting that transient event as evidence here. This comment concerns the sustained burn rate represented by the Usage page: approximately 290M displayed tokens corresponding to about 70% of the weekly allowance.

    Could maintainers correlate the server-side ledger for September 9 and clarify:

    1. the exact model/reasoning weighting applied to the displayed 290M tokens;
    2. the cached-input, uncached-input, output, retry, compaction, subagent, and background-work attribution;
    3. whether the approximately 70% weekly debit was intended under the applicable GPT-6 weighting; and
    4. whether a per-task or per-request usage breakdown can be provided.

    Screenshots or exact client-version details can be added if recoverable. Private task transcripts cannot be posted publicly.

  4. omggga commented on Sep 11, 2026

    @omggga
    omggga
    on Sep 11, 2026

    Same here. The most funny thing is this graph:

    Image

    Seems all my weekly limit went to.. Luna. Beacuse of high Astra capacity. Weird.

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

    CLIIssues related to the Codex CLIbugSomething isn't workingrate-limitsIssues related to rate limits, quotas, and token usage reportingwindows-osIssues related to Codex on Windows systems

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions