Codex Desktop consumes entire 5-hour allowance in ~20 min — 6.4M input tokens / 69 model requests from ~5 user messages #46816

Codex Desktop: 5-hour usage exhausted in ~20 minutes

Codex Desktop is consuming my 5-hour allowance extremely quickly during very small tasks.

On September 20, 2026, I used Codex Desktop on macOS to create a simple microsite. The conversation contained only about 5 user messages and lasted roughly 20 minutes, yet the 5-hour allowance was effectively exhausted.

Local Codex telemetry for the session reports:

• 69 model requests
• 6,422,721 input tokens
• 6,202,112 cached input tokens
• 26,731 output tokens
• 96.6% of input was cached
• context grew from approximately 30k to 141k tokens per request

Near the end of the session, Codex was making repeated requests every few seconds with roughly 140k input tokens each, almost all of them cached.

Example final request:

input_tokens: 140,647
cached_input_tokens: 140,160

Codex-reported cumulative total:

{
"input_tokens": 6422721,
"cached_input_tokens": 6202112,
"output_tokens": 26731,
"reasoning_output_tokens": 3612,
"total_tokens": 6449452
}

This was a new, short conversation, not a long existing thread.

I understand that the 5-hour limit represents compute usage rather than five hours of wall-clock time, but this level of consumption seems abnormal and has started only recently.

Could you please check:

1. whether the usage deducted from my account was calculated correctly,
2. whether cached context is currently being accounted for correctly,
3. whether there is a regression causing Codex Desktop to repeatedly replay large contexts,
4. whether the 69 model calls for such a small task are expected.

I can provide the original Codex .jsonl session logs if needed.