Codex Crashes, Repeated Thinking Failures, and Unexpected 30% Usage Increase

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

26.707.8168.0_x64

What subscription do you have?

x20 Pro

What platform is your computer?

Win 11

What issue are you seeing?

Hello,

I am experiencing serious reliability and usage-accounting issues with the Codex desktop application.

I worked on the same project across three different sessions. My total actual working time was approximately two hours at most, generally in sessions of around 30 minutes.

During this period, I used:

SOL Ultra approximately three times
SOL Medium approximately five times
Then switched back to SOL Ultra

While using Codex, several requests failed with messages such as “Thinking failed.” Some requests did not produce a usable response at all, and the application also crashed.

After the crash, I stopped using Codex and returned approximately four hours later. When I reopened it, my usage had suddenly increased by an additional 30%, despite there being no active usage from me during that period.

The usage page also appeared to show approximately four and a half hours of work, which does not match my actual active usage.

It appears that one or more of the following may have occurred:

Failed “Thinking” attempts were still counted against my usage
Crashed or disconnected sessions continued running in the background
Repeated retries or model-processing attempts were counted even though no successful result was delivered
Usage reporting was delayed and then incorrectly added as a large amount later

The Windows Event Viewer shows the following crash details:

Application name: ChatGPT.exe
Application version: 0.0.0.0
Faulting module name: unknown
Exception code: 0xc06d007f
Fault offset: 0x00007ffca6a51b6a
Process ID: 0x13598
Application path: C:\Program Files\WindowsApps\OpenAI.Codex_26.707.8168.0_x64__2p2nqsd0c76g0\app\ChatGPT.exe
Report ID: a58d3d55-b7c7-4838-8c67-029519494cd2
Package name: OpenAI.Codex_26.707.8168.0_x64__2p2nqsd0c76g0

Could you please review the detailed usage records for my account, including:

The exact requests that consumed the additional 30%
Whether failed “Thinking” attempts consumed usage
Whether any sessions remained active after the application crashed
Whether automatic retries were charged multiple times
The exact model and processing duration associated with each usage entry

I am subscribed to the $200 plan, and losing approximately 30% of the weekly limit after only a small number of requests—many of which failed—is not reasonable.

Please investigate the issue and restore any usage that was consumed by failed requests, application crashes, background sessions, or incorrect accounting.

Thank you.

What steps can reproduce the bug?

Open the Codex desktop app on Windows.
Open the same project in three separate sessions.
Use SOL Ultra for approximately three requests.
Switch to SOL Medium and send approximately five requests.
Switch back to SOL Ultra.
Continue working for a total active time of no more than approximately two hours.
During this process, observe that some requests fail with a “Thinking failed” message and do not return a usable result.
In some cases, the Codex desktop application crashes completely.
After the crash, do not use Codex for approximately four hours.
Reopen the Codex app and check the usage page.

Expected result:
Only successful requests and actual processing time should be counted. Failed requests, crashed sessions, and inactive time should not consume additional usage.

Actual result:
The usage suddenly increases by approximately 30%, and the usage page shows around four and a half hours of work, even though the actual active usage was approximately two hours or less.

What is the expected behavior?

The Codex desktop app should remain stable and should not crash during normal use or when switching between SOL Medium and SOL Ultra.

Requests that fail with “Thinking failed” and do not return a usable result should not consume the same amount of usage as a successfully completed request.

After the application crashes or is closed, all related sessions and background processing should stop. Inactive time should not be counted as active work.

The usage page should accurately reflect the actual requests and processing performed. It should not show approximately four and a half hours of work or consume an additional 30% of the weekly limit when the actual usage was approximately two hours or less.

Any usage caused by failed requests, automatic retries, crashed sessions, or background processes should either not be charged or should be automatically restored.

Additional information

I am using the Codex desktop app on Windows with a $200 subscription plan.

The issue occurred while working on the same project across three separate sessions. My total active working time was approximately two hours or less. I used SOL Ultra for around three requests, SOL Medium for around five requests, and then switched back to SOL Ultra.

Several requests failed with “Thinking failed” and did not return a usable result. The desktop application also crashed. I then stopped using Codex for approximately four hours. When I returned, the usage page showed around four and a half hours of activity and an additional 30% of my weekly limit had been consumed.

This strongly suggests that failed requests, automatic retries, crashed sessions, or background processing may have continued to consume usage after the app stopped responding.

I have also included the Windows Event Viewer crash details in the report. Please review the exact request-level usage records for my account and restore any usage that was consumed incorrectly.