How much are Codex resets worth?

In our 12 days on ChatGPT Pro $100, seven resets took us from a ceiling of about $351 of API-priced Codex to $1,211, roughly 71% of the total value.

How many Codex weekly windows can you get in 12 days?

Normally two at most, because a weekly window lasts 7 days. With two global resets and five banked resets we opened eight.

Should I count on Codex resets when choosing a plan?

No. OpenAI does not promise them and gaps between resets have reached 67 days. Size the plan on the no-reset value and treat resets as a bonus.

Codex resets turned 12 days of a $100 plan into $1,211 of usage

Updated October 9, 2026 · first published October 9, 2026

Quick answer: Twelve days of ChatGPT Pro $100 costs about $40. In those 12 days we pulled $1,211 of API-priced Codex out of it. The plan did not do that. The resets did. Without them, the most 12 days could have...

Twelve days of ChatGPT Pro $100 costs about $40. In those 12 days we pulled $1,211 of API-priced Codex out of it. The plan did not do that. The resets did. Without them, the most 12 days could have held is two weekly windows, about $351. With two global resets and five banked resets, we got eight.

Return on the plan: about 30x the prorated cost.
Share of value that came from resets: about 71%.
Fastest window: 7.2 hours from 0% to 100%.

The gauge, hour by hour

This is the Codex weekly gauge for every hour we used it. Each drop to zero is a reset. B marks a banked reset we redeemed; G marks a global reset OpenAI pushed to everyone. A normal week would show one slow climb. Ours shows eight.

Codex weekly gauge, hour by hour, Sep 27 to Oct 8 (B = banked reset, G = global reset)

Note the drop on October 2. We redeemed a banked reset at 08:49, used 14% of it, and twelve hours later a global reset replaced it. That cost us about $150 of value, and it is why we wrote a guide to when to redeem a banked reset.

What the plan would have delivered without resets

A Codex weekly window starts with your first request and lasts 7 days. Our first Pro window opened on September 27 at 13:26 UTC, so with no resets the second would have opened on October 4 and the 12 days would have ended partway through it. At our measured $175 per full window, the ceiling is about $351, and that assumes we used every bit of both.

Cumulative API value: with resets vs the most a no-reset plan could hold vs what the plan cost

Scenario	Windows in 12 days	API value	Per $1 of plan
Pro $100, no resets (ceiling)	2	$351	$9
Pro $100, the resets we used	8	$1,211	$30
Plus, no resets (for scale)	2	$134	$17

How the math works

Codex writes a JSONL log of every session under ~/.codex/sessions/. Each request records cumulative input, cached input and output tokens, plus the weekly gauge (used_percent) and its resets_at time. We priced every request's token delta at the API list price of the model it ran on and grouped requests by weekly window. Dollar figures are API-equivalent value, not money we paid.

The prorated cost is $100 a month times 12/30, or $40. The no-reset ceiling is two full windows at our median of $175. The hourly series is in codex-pro-100-hourly.csv (gauge percentage plus cumulative API value, cached and uncached) and the windows in codex-pro-100-windows.csv.

Can you count on resets?

No. Public trackers have logged 59 resets since September 2025, and the pace sped up sharply this summer: 12 in July, 10 in August, 7 in September. But OpenAI has never promised them, and the gap between resets has been as long as 67 days. Price your plan on the no-reset line and treat everything above it as a bonus. Our Codex reset log has the dated list.

The multiplier matters too. A full Pro $100 window held 2.6x a Plus week, not 5x (details here). Resets are what made the plan a bargain, not the plan.

Sources

Want this applied to your stack? Bring the provider bills, gateway logs, and top workflows; we will map the cost drivers and savings path. Book a free audit →

Back to finopsllm.com