Codex usage limit is reached too quickly, even with ChatGPT Plus

Codex
1

I am hitting Codex usage limits after only a few prompts in the Visual Studio extension, and even with ChatGPT Plus and purchased credits, the available usage feels too limited for real coding work.

I have been using OpenAI tools for a long time, and I am currently using ChatGPT Plus, but I am facing a repeated issue specifically with Codex in the Visual Studio Code extension.

The problem is that the usage limit gets exhausted very quickly — sometimes after only 3–4 prompts. For actual development work, this is too restrictive, because even a small coding task often needs multiple iterations, refinements, and follow-up prompts.

What is also confusing is that even after buying credits, the practical usage still feels very limited. Because of this, it is not clear:

  • why Codex usage runs out so fast,

  • how usage is being calculated in the VS Code extension,

  • why purchased credits do not seem to meaningfully extend working time,

  • and what plan or option is actually suitable for developers who rely on Codex for regular coding sessions.

Since I already have ChatGPT Plus, I would really appreciate clarification on how Codex limits work separately from ChatGPT, and whether there is a better way to get a more usable coding workflow without hitting limits after just a few prompts.

Is anyone else facing the same issue with Codex in VS Code?

2

Yes, I’m seeing the same issue.

I’m also using Codex (via the CLI), and the usage runs out very quickly — even with just a few prompts. It makes it hard to use for real development work where you need multiple iterations.

This really needs to be fixed as soon as possible. Right now it feels like we’re paying for tokens that we can’t effectively use, which is not acceptable at all.

3

It is because they recently changed the way usage is consumed and it’s become much more limited as a result. It’s no longer averaged per message, but regular token based pricing like the API. If you want it to last longer ,you’d need to use a lower reasoning or model.

https://help.openai.com/en/articles/20001106-codex-rate-card

It sucks, but I hope it helps you understand.