Codex Pro 5h quota drops 2% after trivial prompts and capacity errors #30034

What version of Codex CLI is running?

v0.142.2

What subscription do you have?

ChatGPT Pro

Which model were you using?

gpt-5.5, reasoning medium, summaries auto

What platform is your computer?

Windows x64

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

Windows Command Prompt (cmd.exe)

Codex doctor report

not available

What issue are you seeing?

I’m seeing what looks like abnormal Codex Pro 5-hour quota drain in Codex CLI.

My Codex Pro 5h limit dropped from 98% left to 96% left after only a few trivial prompts such as 1+1, 你好, and 2+2, plus several Selected model is at capacity failures.

There was no repository analysis, no file editing, no file reading, no long-running agent task, and no meaningful coding work during this test.

What steps can reproduce the bug?

1. Start Codex CLI on Windows:

codex

2. Run /status.

Initial status showed:

Codex CLI version: v0.142.2
Model: gpt-5.5 (reasoning medium, summaries auto)
Account: <redacted> (Pro)
5h limit: 98% left
Weekly limit: 77% left

The CLI also showed:

You have 1 usage limit reset available. Run /usage to use one.

I did not use /usage.

3. Send only trivial prompts, with no repository analysis, no file reads, no file edits, and no long-running agent task:

1+1等于几

Codex replied:

2

4. Run /status again.

Status still showed:

Context window: 97% left (19.2K used / 258K)
5h limit: 98% left
Weekly limit: 77% left

5. Send a few more trivial prompts:

1+1等于几
2+2等于几
你好
你好呀
1+1等于几
你好
2+2等于几

Several requests failed with:

Selected model is at capacity. Please try a different model.

Successful responses were only very short answers, such as:

2
你好。
4

6. Run /status again.

Final status showed:

Context window: 97% left (19.3K used / 258K)
5h limit: 96% left
Weekly limit: 77% left

The 5-hour quota dropped from:

98% left -> 96% left

The weekly quota stayed at:

77% left

No coding task was performed during this test.

What is the expected behavior?

Trivial prompts such as 1+1, 你好, and 2+2 should not consume 2% of the Codex Pro 5-hour quota.

Requests that fail with:

Selected model is at capacity. Please try a different model.

should not appear to cause excessive quota drain.

If capacity failures are counted against quota, the CLI should make that behavior clear. If they are not supposed to be counted, then this looks like a quota metering or reporting bug.

Additional information

Environment:

Codex CLI version: v0.142.2
OS: Windows x64
Subscription: ChatGPT Pro
Model: gpt-5.5
Reasoning: medium
Summaries: auto
Directory: ~
Permissions: Workspace (Ask for approval)
Collaboration mode: Default

I understand that Codex CLI requests may include system prompts, tool instructions, session state, and previous context. The /status output showed around 19.2K–19.3K used / 258K context.

However, the 5-hour quota drain still seems excessive for such a minimal interaction, especially because there was no codebase analysis, no file reading, no file editing, and no long-running agent task.

This may be related to one of the following:

• 5-hour quota metering issue
• Failed/capacity requests being counted against quota
• Incorrect model pool accounting for gpt-5.5
• Codex CLI reporting incorrect usage
• Account-side quota state issue after reset

Could the team please check whether this is expected behavior, a quota metering issue, or a CLI reporting bug?