Codex usage transparency and a subscription-based API – am I missing something?

[Area51FR]

I've been using Codex more and more for actual development work, and there are two things that keep bothering me.

Maybe I'm missing something obvious, but I'd be interested to hear how other people see this.

1. I really want to know what a task is going to cost me

When ChatGPT switches a development task over to Work/Codex, I currently have very little feeling for how much of my available usage that task is going to consume.

That's especially annoying when choosing between models.

I don't necessarily need a perfectly accurate token prediction. Even a rough estimate would already be extremely useful:

Estimated usage: ~2–5% of your remaining allowance
Model: Sol Light
Complexity: Medium

And if I select a more expensive model:

Astra6 is expected to use approximately 2–3x more of your allowance for this task.

Then, while it's running, show the actual consumption. And when it's finished:

Estimated: 2–5%
Actual: 3.7%

That would teach me very quickly which model makes sense for which type of development task.

Right now I can see that my allowance is disappearing, but I can't really associate that consumption with individual tasks. For a paid account, that feels unnecessarily opaque.

Interestingly, Codex is very careful about asking for permission to perform actions. That's understandable, of course. But once I've deliberately enabled Full Access, I'd personally find a usage warning much more valuable than another permission dialog.

Tell me what you're going to consume. I'll decide whether the task is worth it. :-)

2. Why can't my ChatGPT/Codex allowance also be used through an API?

This is the other thing I don't quite understand.

I'm already paying for a ChatGPT account that includes Codex/Work usage.

So why should the interface determine how that usage is billed?

If I use Codex through ChatGPT, the Codex app, CLI or an IDE, I'm consuming my subscription allowance.

But if I want to build my own small development tool around the same models through an API, suddenly I'm in a completely separate billing world.

I'd love to see something like a ChatGPT-authenticated Codex API.

Not a "free API" in the sense of unlimited free tokens.

Simply an API that consumes the same allowance I've already paid for.

Something conceptually as simple as:

My App → Codex API → ChatGPT authentication → my existing Codex/Work allowance

If I have 42% of my allowance remaining and my API task consumes 1.4%, afterwards I have 40.6%.

Simple.

To me, usage is usage.

Whether those tokens were requested by the ChatGPT UI, Codex Desktop, CLI, VS Code or a little tool I wrote myself shouldn't fundamentally change the economics.

And combining both ideas would make this even better:

One account. One agentic allowance. Multiple interfaces. Transparent consumption.

Before a task:

estimated usage

During a task:

actual consumption

After a task:

final usage

And it shouldn't matter whether the task originated from ChatGPT, Codex, CLI, IDE or an authenticated API.

For developers in particular, I think this would make Codex much easier to integrate into our own workflows while also making us much more conscious of what we're actually consuming.

Maybe there are technical, security or commercial reasons why OpenAI keeps these things separate today. I'd genuinely be interested in understanding them.

But from the user's side, the current separation feels rather artificial.

What do you think?

Would you use a Codex API tied directly to your ChatGPT allowance? And would a per-task usage estimate change how you choose between models?