[Windows][Pro] Critical Codex performance degradation, stalled agents and repeated HTTP 429 errors #54004

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

ChatGPT Powered by Codex & OWL Verzia 26.1007.71326

What subscription do you have?

ChatGPT Pro 200 (€229/month), signed in with ChatGPT)

What platform is your computer?

Windows 11 Pro (64-bit) 26H2 26300.9711

What issue are you seeing?

Critical and continuously worsening Codex performance

I am a paying ChatGPT Pro 200 subscriber (€229/month), and I am experiencing extremely serious and progressively worsening performance and reliability issues with Codex.

Over the last several days, Codex has become practically unusable for professional software development.

The problems include:

• Extremely slow response generation and agent execution.
• Prolonged periods of inactivity without meaningful progress.
• Agents appearing stuck for extended periods.
• Coding tasks taking significantly longer than they previously did.
• Repeated interruptions and failed executions.
• Unpredictable execution times, even for ordinary development tasks.
• Repeated HTTP 429 errors.

The exact error message is:

exceeded retry limit, last status: 429 Too Many Requests

The affected models include GPT-6 Astra and GPT-6.1 Sol, including configurations using high and maximum reasoning effort.

The situation is continuously deteriorating. In my experience, Codex has never been this slow or unreliable.

My current usage dashboard shows:

• Active ChatGPT Pro 200 subscription
• 80% of the weekly allowance remaining
• 60,110 available credits

These figures are from the usage screenshot and do not necessarily represent the allowance at the exact moment of each error.

I understand that additional rate limits may apply. However, the severe performance degradation also occurs when no HTTP 429 error is displayed.

Important: ChatGPT itself is also experiencing extremely severe and progressively worsening response latency. This is not limited to Codex.

I have already escalated the issue through OpenAI Support, and the case has been forwarded to a support specialist.

Please investigate the persistent performance degradation, repeated HTTP 429 errors, backend latency, and potential capacity or request-throttling issues.

This is seriously affecting my professional work and the usability of my paid subscription.

What steps can reproduce the bug?

The issue is intermittent but occurs repeatedly during normal development work.

1. Open the Codex application on Windows.
2. Sign in using an active ChatGPT Pro subscription.
3. Start or resume a software development session.
4. Select GPT-6 Astra or GPT-6.1 Sol with a high or maximum reasoning configuration.
5. Submit a normal coding, debugging, or code-analysis task.
6. Observe extremely long response times, prolonged inactivity, or stalled agent execution.
7. During some sessions, Codex fails with:

exceeded retry limit, last status: 429 Too Many Requests

The severe slowdown also occurs without an explicit error message.

The issue has become progressively worse over recent days.

I cannot reproduce the HTTP 429 error deterministically on every request, but both the latency and error symptoms have occurred repeatedly.

Additional timestamps, affected sessions, and diagnostic logs can be provided privately if required.

What is the expected behavior?

Codex should provide reliable, reasonably responsive execution for professional software development.

Expected behavior:

• Agent tasks should begin processing without excessive delays.
• Progress should continue without unexplained prolonged inactivity.
• Coding and analysis operations should complete within reasonable timeframes for their complexity.
• Sessions should not repeatedly fail with unexplained HTTP 429 errors.
• Available account allowances and applicable rate limits should be reported clearly.
• Temporary backend errors should be handled through appropriate retry mechanisms.
• Performance should remain sufficiently consistent for professional development workflows.

I do not expect every complex task to finish immediately, but the current extreme latency, frequent stalls, and recurring failures are far beyond normal expectations.

I expect the service to return to a stable and usable level of performance.

Additional information

Additional context and request for engineering investigation

Subscription: ChatGPT Pro 200 (€229/month)

Usage at the time of the screenshot:

• 80% weekly allowance remaining
• 60,110 credits available

Affected period: Performance has progressively deteriorated, becoming especially severe in the days leading up to October 14, 2026.

Affected services: Codex and ChatGPT.

I have already submitted a formal complaint to OpenAI Support. The support system confirmed:

"Escalated to a support specialist; You can expect a response in the coming days."

Other users have reported similar issues:

• Codex is becoming unusable: extreme slowness and 'at capacity' errors across all models #53591
• [Windows 10 19045] GPT-6.1 Sol: persistent "at capacity" / HTTP 503 with 94% quota remaining (Oct 12-13), corroborating #52812 / #53407 #53559

I understand that these reports may not share the same underlying cause.

Requested investigation

Please investigate:

1. The cause of extreme model inference and agent execution latency.
2. Repeated HTTP 429 errors despite substantial remaining weekly allowance.
3. Possible backend congestion, request throttling, or capacity constraints.
4. Whether higher reasoning configurations are experiencing disproportionate performance problems.
5. Whether request retries are functioning correctly.
6. Whether these problems are part of a broader infrastructure or performance regression.

Please escalate this issue to the Codex engineering and infrastructure teams.

I would appreciate a clear technical explanation, corrective action, and an estimated resolution timeframe.

The current service performance has become unacceptable for professional use and is seriously undermining the value of my premium subscription.

In-app feedback reference

I have already submitted a diagnostic feedback report through the ChatGPT Windows desktop application.

Feedback ID:
no-active-thread-01a13ae2-7e25-71a2-8e53-3a92115e5a96

Diagnostic logs were enabled during submission.

The feedback system returned a "no-active-thread" identifier, so it is unclear whether the report was associated with an active conversation.

Please use this reference to correlate my feedback with the reported performance issues if possible.

I have also escalated the issue through OpenAI Support.