Codex built-in imagegen returns TooManyRequests even though no image is generated

What version of Codex CLI is running?

codex-cli 0.137.0

What subscription do you have?

Pro x20

Which model were you using?

gpt-5.5

What platform is your computer?

Microsoft Windows NT 10.0.26200.0 x64

What terminal emulator and version are you using (if applicable)?

Windows Terminal / PowerShell

Codex doctor report

Run this and paste the full output into the GitHub form:


codex doctor --json


In this environment, `codex doctor --json` completed successfully with:


overallStatus: ok
codexVersion: 0.137.0
auth mode: chatgpt
model: gpt-5.5
network.websocket_reachability: ok
updates.status: current version is not older

What issue are you seeing?

Codex CLI built-in image generation repeatedly returns TooManyRequests for ordinary new image generation prompts.

The failed attempts do not appear to create any image files. On this machine, C:\Home\.codex\generated_images still shows the latest successful generated image as:

2026-06-05 23:42:12 JST

Repeated attempts after that time returned TooManyRequests, but no new generated image directory or file appeared.

ChatGPT image generation appears usable as a workaround, so the problem seems specific to the Codex CLI built-in image generation path or its rate-limit state.

What steps can reproduce the bug?

1. Open Codex CLI on Windows.
2. Ask Codex to generate a new image using the built-in image_gen / image generation tool.
3. Use an ordinary new-image prompt, for example a small desktop-gallery anime illustration.
4. Observe that the tool returns TooManyRequests.
5. Check C:\Home\.codex\generated_images.
6. Observe that no new generated image directory or file was created.
7. Retry with a shorter/lighter prompt.
8. Observe the same TooManyRequests result and no generated output.

What is the expected behavior?

One of these should happen:

• The image is generated successfully.
• Or Codex returns a precise service/degradation error if the hosted image generation tool is unavailable.

Failed attempts that produce no image should not behave like successful quota usage.

Additional information

• Surface: Codex CLI on Windows
• Tool path: built-in image_gen / image generation tool invoked from Codex
• Not using direct OpenAI API
• Not using CLI fallback image generation
• Auth mode: ChatGPT auth
• Timezone: Asia/Tokyo

Could you confirm whether Codex CLI built-in image generation is affected by a known incident, stale rate-limit state, or incorrect TooManyRequests handling?

If this is a platform-side issue, a status note or clearer tool error would help users avoid misdiagnosing it as local quota exhaustion.