How Far Does 500 Credits Go?

I was using Codex, OpenAI’s coding agent, when its usage meter told me I had 7% left.

That number made sense to me.

I’d used Codex enough to know what 7% felt like. I’d hit the usage limits before, watched the allowance come back, and carried on. I knew the broad mechanics of the plan.

I couldn’t have told you exactly how much work one percentage point represented.

I’d just never needed to.

7% meant I was nearly out.

Then I got the option to buy more usage.

500 credits. ₹1,700.

I knew what usage credits were too.

What I didn’t know was what 500 Codex credits meant for the kind of work I’d been doing.

Another afternoon? A few days? Much more?

No idea.

That was what bothered me.

I had a number I understood well enough right up until I had to decide whether to spend money on another number.

So I went looking for a conversion.

Not an exact one. I wasn’t expecting: 500 credits = exactly another 73%

Just something that would tell me whether 500 was a little or a lot.

The Codex rate card made the problem clearer pretty quickly.

A typical GPT-5.6 Sol task can use roughly 5–40 credits. The amount depends on what the task actually involves, including the input, cached input and output.

Which makes sense.

“Fix this typo” and “read this repository, work out what’s wrong, change several files and run through the result” might both look like one request from my side.

They obviously aren’t the same amount of work.

So: 500 credits = another X% wasn’t really the useful conversion I’d hoped for. Fair enough.

That killed the easiest version of my complaint.

The slightly annoying part was what I noticed next.

7% hadn’t actually told me very much either.

It didn’t tell me how many tasks I had left. It didn’t tell me how much the next task would use.

The percentage felt informative because I’d spent time with it.

I’d watched it fall while I worked.

After enough of that, I’d built a rough feel for the number without deliberately learning it.

60% felt fine.

20% felt low.

7% meant I should probably stop wasting it.

There was nothing especially transparent about the percentage itself. I was just calibrated to it.

I wasn’t calibrated to 500 credits.

I could read the definition. I could read the rate card. I could do the arithmetic and discover that 500 divided by a 5–40 credit range gives me an impressively large range of possible tasks.

That still didn’t tell me what 500 would feel like for my use.

And I suppose that was the whole thing.

The first number had become useful through experience. The second arrived with a price attached before I’d had any experience with it.