How should developers plan around Codex resets?

Since the random resets started happening, I’ve been constantly wondering if I have to use up my entire limit over the weekend because Tibo will reset on Monday :folded_hands: , or if I should save up because my reset isn’t scheduled until Friday… I literally have mixed feelings (joy and fear at the same time) – it’s gambler’s syndrome → if I use everything up on intense work over the weekend and get a reset on Monday, I’ll feel happy – and if not, I’ll be sad.

These are just my personal thoughts. :melting_face:

Okay, now we get it in the form of a ticket (we can decide for ourselves when we want) - this is a much better solution. :smiling_face:

:thinking:

You mean like the banked resets?

you can check here: https://chatgpt.com/#settings/Usage

here’s the official instructions: https://help.openai.com/en/articles/20001498-how-banked-codex-resets-work

You can follow Tibo on Twitter to learn in advance about upcoming resets.
Nowadays the date and ~San Francisco time will be announced in advance.

And talking about resets:

@karolplu
I forgot to ask:

Do you have special projects to use the remaining tokens or do you wait for a reset to go into regular work overdrive?

Maybe this is actually about finding more relaxed ways to manage tokens?

Also this thing but have not updated it in a bit: :slight_smile:

What if the goal isn’t to predict every reset, but only the ones that seem almost certain?

There will most likely be a reset for DevDay 2026 on September 29, and I would expect another one when Codex reaches 30 million users.

well now there is bonus banked resets to wrestle with as well.

I’m currently designing tools for drawing technical drawings of fences - connected to synchronized data and a configurator. - But the codex is now an everyday tool for me.

But from the very beginning, I’ve been modifying Codex settings to optimize the usage and performance associated with subagents.

My Codex setup is optimized as an autonomous senior developer and QA engineer:

Sol XHigh is the default model; I select the main model manually.

Subagents are used only for complex tasks: XHigh up to 2, Max up to 3, Ultra up to 8.

Codex selects Luna or Terra based on task complexity, reuses existing agents, and displays their model in the name.

A hard hook enforces agent limits, model selection, and prevents subagents from creating more agents.

Project memory, handoffs, and SessionStart restore context after restarts or compaction.

The Visualizer project has 828.4K usable context tokens.

Skills and plugins are automatically selected for WordPress, coding, UI/UX, testing, and visual QA.

Codex autonomously analyzes, implements, tests, and commits changes, but asks before production operations, paid actions, data deletion, or git push.

Unfortunately, sometimes (quite often) I hit the limit.

Therefore, to be able to work smoothly, I need to know how much Codex power I can utilize. With random restarts, I often felt like I was missing out on a chance (I’m not complaining - I understand how it works), and I have to say that the current distribution of resets (I decide when to use them) is simply healthy.

But I have a simple idea… next to the use icon, add a symbolic number for any unused resets (the number indicates how many resets you have left – if you don’t have any, the number isn’t displayed).

Also, when you launch Codex, when you receive a new reset, display annotations and a short comment explaining the occasion you received the reset, so you can use it at any time.

I think many people will appreciate this.

A new reset is available!
To celebrate this new milestone, we're giving you a little gift. Enjoy!

I suspect many people don’t know they have this option.
If I hadn’t followed this forum, I wouldn’t have found out about Tibo…
¯\_(ツ)_/¯

Welcome to the site, @karolplu … Glad you found us! You can read about the history of this community if you have the time. Very interesting read, imho.

It’s great to hear ideas from people who are heavily using Codex. Hope you stick around!