$200 Pro weekly allowance repeatedly exhausted in about two days with GPT-6 Astra Extra High

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

Desktop app: 26.908.40834 Codex runtime: 0.154.0-alpha.6.2

What subscription do you have?

ChatGPT Pro individual subscription, USD 200/month.

What platform is your computer?

Microsoft Windows NT 10.0.26200.0 x64 Windows 11, build 26200.8655 Codex desktop app; local Windows workflow with PowerShell.

What issue are you seeing?

I am a ChatGPT Pro subscriber paying USD 200 per month. My weekly Codex allowance repeatedly lasts only approximately 2–2.5 calendar days when my main working model is GPT-6 Astra at Extra High. In my latest experience, it lasted about two days.

This leaves me unable to continue with the same model and settings for the remainder of the weekly window unless I use an available reset or purchase additional credits.

The problem is not simply that I am running a capable model. A recurring part of my workflow consists of correcting errors introduced by the agent itself: failed fixes, incorrect test preparation, repeated validation, reconstruction of task context, and further repairs. These activities consume usage while the intended end-to-end result can remain unfinished.

A recent review of saved conversation and execution records confirmed repeated agent-caused repair cycles. However, I cannot determine exactly how much allowance each failure consumed. I am not claiming that all usage was wasted or that a billing error has already been proven.

My main concerns are:

1. The included allowance does not provide predictable working-week capacity for my actual development workflow.
2. I cannot adequately attribute the weekly percentage consumed to individual tasks, reasoning settings, speed settings, retries, or other execution overhead.
3. Repeated unsuccessful repairs can consume substantial capacity before the agent delivers a completed, verified result.
4. Buying more credits does not address the underlying problem when the additional capacity is spent repeating unsuccessful work.

I understand that a weekly allowance is a budget, not a guarantee of seven days of continuous execution. Nevertheless, repeatedly exhausting a USD 200 professional subscription in about two days makes it difficult to rely on Codex for ongoing work.

Please investigate whether this consumption is expected under the current allowance and model pricing, whether abnormal repeated execution contributes to it, and what concrete options exist for sustainable Astra Extra High usage at standard speed.

What steps can reproduce the bug?

This describes a recurring observed usage pattern, not a deterministic minimal reproduction. I am not asking anyone to repeat an expensive workload solely to reproduce quota exhaustion.

1. Use the Codex desktop app on Windows with a USD 200/month ChatGPT Pro subscription.
2. Use GPT-6 Astra, primarily at Extra High, for ongoing local software development.
3. Continue tasks through implementation, testing, corrections, and verification.
4. During unsuccessful tasks, observe repeated repair attempts, test-preparation corrections, or restoration of previously established task context.
5. Check the account's weekly usage indicator throughout the allowance window.
6. In my experience, the weekly allowance is exhausted after approximately 2–2.5 calendar days, with the latest experience being about two days, while some intended outcomes remain unfinished.

These are elapsed calendar days, not a measured number of continuous model-execution hours.

Please use account-side records to verify the exact timestamps, models, reasoning levels, speed settings, concurrent activity, and consumption. Session references can be supplied privately. An exact billed-token count and per-task share of the weekly allowance are not established in this report.

What is the expected behavior?

I would like a predictable and sustainable professional workflow

Specifically, I am requesting:

1. A clear explanation of the workload the USD 200 Pro allowance is intended to support with Astra Extra High, and why my allowance repeatedly lasts only about two days.
2. Transparent usage attribution by task, model, reasoning level, and speed, with an explanation of how retries and other execution overhead affect consumption.
3. Consideration of a larger included allowance or an optional slower, lower-cost mode that preserves the requested model and reasoning quality.
4. Configurable per-task spending limits and early warnings when repeated unsuccessful repair attempts are consuming substantial capacity.
5. Better reuse of valid prior work and earlier detection of repair loops, without skipping necessary verification.
6. A private review of my affected sessions, with consideration of an allowance restoration or credit adjustment if platform failures or abnormal repeated execution are confirmed.

Please distinguish between expected expensive usage, confirmed technical failures, and potentially avoidable execution overhead. If this consumption is expected, please explain what practical configuration or allowance would support my workflow throughout a working week.

Additional information

That report concerns repeated repair and replanning cycles. This new issue focuses specifically on the practical capacity of the USD 200 Pro subscription: the weekly allowance repeatedly running out after about two days, the lack of clear attribution, and the cost of unsuccessful repair cycles.

My usual model setting is GPT-6 Astra Extra High. The reviewed historical records were not an exclusively Extra High sample: one

My request is for sustainable Extra High usage at standard speed, without requiring Fast mode or Ultra.

No repository file, test, or code entry point is identified. Start by reviewing the account-side records and session references mentioned in the report, comparing timestamps, models, reasoning levels, speed settings, concurrent activity, and consumption; done means distinguishing expected usage from technical failures or abnormal repeated execution.