Pro Lite 5x: GPT-5.5 / GPT-5.4-mini return 429 with 12% weekly remaining, Spark works

What issue are you seeing?

I am using OpenAI Codex CLI v0.136.0 on macOS.

Account:

• zhengkaiweics@gmail.com
• Plan shown in /status: Pro Lite

/status:

• Model: gpt-5.5, reasoning high
• 5h limit: 39% left, resets 23:50
• Weekly limit: 12% left, resets 08:22 on 7 Jun
• GPT-5.3-Codex-Spark 5h limit: 100% left
• GPT-5.3-Codex-Spark weekly limit: 100% left
• Directory: /private/tmp/codex-test
• Codex CLI version: v0.136.0

Problem:
In an empty test directory, even a trivial prompt fails on the main Codex quota pool.

Example 1:
Model: gpt-5.5, reasoning high
Prompt: hi
Error:
Falling back from WebSockets to HTTPS transport. unexpected status 403 Forbidden:
url: wss://chatgpt.com/backend-api/codex/responses
cf-ray: a05c33632f2d0653-SJC

exceeded retry limit, last status: 429 Too Many Requests
request id: a6a6bc46-1364-4e5c-8a3c-c9fee0124185

Example 2:
Model: gpt-5.4-mini
Prompt: hi
Result: same 429 behavior.

Control test:
After switching to gpt-5.3-codex-spark high, the same prompt works:
Prompt: hi
Response: "Hi. What would you like me to work on in /private/tmp/codex-test?"

Expected:
With 12% weekly quota remaining on a Pro Lite / 5x account, trivial prompts should still work, especially on mini models. If the remaining quota is below the minimum threshold for some models, /status should report that clearly instead of showing 12% left.

Actual:
gpt-5.5 and gpt-5.4-mini return 429 even for "hi", while Spark works.

This suggests a quota gate, usage reporting, or model-pool accounting bug.

What steps can reproduce the bug?

No

What is the expected behavior?

_No response_

Additional information

_No response_