# Burning through my $100 Codex plan in 2 days. Any advice? What about free resets ? Also considering Claude Code — retained record

> Retained reddit record from reddit:ChatGPT.

Canonical: https://highsignal.app/data/records/d0edaadd3c7e0f56

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ChatGPT)  published 2026-11-03 09:00 ingested 2026-11-04 02:37

[original source ↗](https://www.reddit.com/r/ChatGPT/comments/1xmo1r0/burning_through_my_100_codex_plan_in_2_days_any/)

## retained text



```
Edit : Thanks a lot everyone for your advice. It's gonna help a lot !!  
  
Hello everyone!

I'm kinda new to Codex and advanced AI stuff. I'm currently trying to build a website that requires quite a lot of work (it's a complete tool with its own engine and everything). It'd probably be easy for you to develop it, but I'm not a dev, just a random guy trying to bring an idea to life, so I feel like I'm missing some pretty important knowledge to get things done properly

Right now, we're still struggling to stabilize the engine and its architecture. Fixing one thing sometimes breaks another, some parts of the system don't behave consistently, and we're trying to move away from having lots of individual fixes towards something more unified and reliable. So it's a lot of structural work, not just building pages or adding features.

**My problem is that 2 days of (intense) work basically burn through my entire $100 plan.**

I use Astra mostly to plan workflows, and occasionally for particularly tricky tasks. For pretty much everything else, I use Sol 6.1.

I tried asking Astra to design a less token-hungry workflow, using Terra and Luna for some tasks, but it didn't go very well. It feels like once a model starts working on something, it's better to let it handle the whole task, otherwise things get messy pretty quickly.

And the problem is that almost none of my current tasks seem suitable for cheaper models. Maybe that's just because we're still working on the engine and architecture, and things are too interconnected at this stage?

Anyway, I have two questions:

**1) I'm thinking about trying Claude Code too.**

I'm hesitating between upgrading to the $200 ChatGPT plan, or keeping my $100 plan and getting a $100 Claude subscription on top of it.

I'd like to see which one works better for my project, but I'm also a bit afraid that switching between the two might create an even bigger mess.

I've noticed there have been quite a few free quota resets over the past few weeks. Was that something special, or do you generally get something like one free reset per week?

If those resets are a regular thing, I might just stick with Codex, as that would probably be enough for me to finish the project. If not, I figured I might as well give Claude a try.

**2) How do you guys manage complex projects without burning through so many tokens?**

Do you have specific workflows, ways of splitting tasks, or tricks to reduce consumption? Especially when working on an interconnected codebase where cheaper models tend to make mistakes?

I feel like I'm spending a lot of tokens just trying to stabilize things, sometimes going back and forth on the same issues. I'm not sure how much of that is normal for this kind of project, and how much comes from me not knowing how to manage AI development properly.

Any advice would be really appreciated :)
```



document fetched 2026-11-04 00:20 · canonical https://reddit.com/r/ChatGPT/comments/1xmo1r0/burning_through_my_100_codex_plan_in_2_days_any
