Usage limits draining abnormally fast on Plus: 70% to 100% in about 6 minutes on July 2, 2026

[gititya]

Summary

Codex usage accounting appears to drain the 5-hour limit abnormally fast on ChatGPT Plus, especially on July 2, 2026. Based on local Codex session logs on macOS, the primary 5-hour usage jumped from 70% to 100% in about 5 minutes 48 seconds during ordinary interactive use, and another local session shows 71% already consumed before the visible prompts in that session began.

This does not look like normal model usage. It looks like either:

• incorrect usage accounting,
• hidden/background usage attribution,
• or a serious regression in per-turn weighting.

Environment

• Platform: macOS
• Plan: Plus
• App: Codex Desktop / VS Code source
• Observed models in affected sessions:
  • gpt-5.5 with medium effort
  • another July 2 investigation session on gpt-5.4 with medium effort also continued draining rapidly

Strongest evidence

Post-reset timeline on July 2, 2026

From local Codex usage log events after the reset:

• 2026-07-02 15:18:21 IST -> 1%
• 2026-07-02 15:18:30 IST -> 2%
• 2026-07-02 15:18:46 IST -> 3%
• 2026-07-02 15:19:21 IST -> 6%
• 2026-07-02 15:19:41 IST -> 7%
• 2026-07-02 15:20:00 IST -> 9%

Then later in the same local session history:

• 2026-07-02 15:40:11 IST -> 70%

So the local logs show a post-reset climb from 1% to 9% within about 1 minute 39 seconds, followed by a jump to 70% within about 21 minutes 50 seconds. I am not attaching raw session files here because they contain prompts, outputs, and local workspace details. I can provide sanitized token_count / rate_limits lines only if needed.

Session 019f1ee9-693d-7251-bd45-953c7de51d84

Local session file:

• ~/.codex/sessions/2026/07/01/rollout-2026-07-01T23-50-41-019f1ee9-693d-7251-bd45-953c7de51d84.jsonl

This session is explicitly recorded as gpt-5.5 with medium effort.

Primary 5-hour usage timeline:

• 2026-07-01 23:51:00 IST -> 5%
• 2026-07-01 23:51:04 IST -> 7%
• 2026-07-01 23:52:49 IST -> 15%
• 2026-07-01 23:53:44 IST -> 20%
• 2026-07-02 15:18:21 IST -> 1%
• 2026-07-02 15:18:30 IST -> 2%
• 2026-07-02 15:18:46 IST -> 3%
• 2026-07-02 15:19:21 IST -> 6%
• 2026-07-02 15:19:41 IST -> 7%
• 2026-07-02 15:20:00 IST -> 9%
• 2026-07-02 15:40:11 IST -> 70%
• 2026-07-02 15:40:22 IST -> 72%
• 2026-07-02 15:42:05 IST -> 77%
• 2026-07-02 15:42:39 IST -> 80%
• 2026-07-02 15:43:13 IST -> 84%
• 2026-07-02 15:43:56 IST -> 87%
• 2026-07-02 15:44:58 IST -> 95%
• 2026-07-02 15:45:37 IST -> 98%
• 2026-07-02 15:45:59 IST -> 100%

That is a rise from 70% to 100% in about 5 minutes 48 seconds.

Session 019f224e-b04a-7ee2-a2ab-b61b66f3c944

Local session file:

• ~/.codex/sessions/2026/07/02/rollout-2026-07-02T15-40-10-019f224e-b04a-7ee2-a2ab-b61b66f3c944.jsonl

This session shows:

• 2026-07-02 15:42:07 IST -> 77%
• 2026-07-02 15:43:16 IST -> 85%
• 2026-07-02 15:43:25 IST -> 86%
• 2026-07-02 15:43:33 IST -> 87%

Important detail: another local session file showed the early post-reset 1%-9% steps above, but this session history is where the later 70%-100% drain is clearly visible.

Session 019f17cd-bde4-7802-8e91-07528398fb02

Local session file:

• ~/.codex/sessions/2026/06/30/rollout-2026-06-30T14-43-07-019f17cd-bde4-7802-8e91-07528398fb02.jsonl

This earlier session shows a much more gradual pattern at the beginning:

• 2026-06-30 14:46:27 IST -> 3%
• 2026-06-30 14:46:33 IST -> 4%
• 2026-06-30 14:46:46 IST -> 5%
• 2026-06-30 14:47:06 IST -> 6%
• 2026-06-30 14:47:50 IST -> 7%
• 2026-06-30 14:48:16 IST -> 8%
• 2026-06-30 14:48:31 IST -> 9%
• 2026-06-30 14:48:36 IST -> 10%
• 2026-06-30 14:49:13 IST -> 11%
• 2026-06-30 14:49:19 IST -> 12%
• 2026-06-30 14:49:33 IST -> 13%

This makes the July 2 behavior stand out as abnormal.

Why this looks broken

• A local gpt-5.5 session rose from 70% to 100% in under 6 minutes.
• Another July 2 local session already started with most of the 5-hour budget gone, with no local session artifact from 15:25 IST to 15:40 IST explaining that jump.
• The earlier June 30 session shows a much more incremental consumption pattern.
• OpenAI publicly said a Codex usage-drain fix had been deployed on June 30, 2026, but this July 2 behavior still appears broken.

Ask

• Can you investigate whether usage accounting is overcharging or attributing hidden/background work to the 5-hour limit?
• Can you confirm whether per-turn weighting changed for Plus users around July 1-2, 2026?
• If useful, I can provide the exact local JSONL session files and the token_count / rate_limits events from those files.

[github-actions]

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

• Codex Pro 5x quota drops 2-3% for a single normal prompt, equivalent to 10-15% of Plus 5-hour quota #30909
• Token drain issue w/ multiple agents #30842
• Extreme Tokens usage for similar tasks #30694
• Excessive consumption of the limit: -13% for the "ping" message #30357
• Codex (gpt-5.5, Plus plan) — App is draining usage limits without any activity on the app #29587

Powered by Codex Action

[Katalyst-Sports]

I just gave codex a prompt and it ran for 30 minutes, used up all credits for the 5 hour window. Something is definitely broken.

[Katalyst-Sports]

I just gave codex a prompt and it ran for 30 minutes, used up all credits for the 5 hour window. Something is definitely broken.

019f14b6-e3eb-79c1-8e55-de86fe29e7fc

[tripplyons]

I'm also seeing my usage being consumed at a faster rate today than before today.

[janzofx]

Have the same problem

• Codex 100 $ plan - 2x 3 hours session on 5.5 high/medium reasoning, 48 % of weekly limit gone.
• Enterprise plan - 4x 5h window = weekly usage gone
• Easy task - codex working 1 hour in loop, editing same files 10 times and didn't even fix it, eats half of 5 hour window, today this happened 4 times to me
• You write something in the prompt to do that that way - codex will decide to do it completely different, you try to prompt again to do it that way, again no success - this I saw in the last 2 weeks most.

I am working with codex 10-12 hours a day, so these are not things that I saw one or two times