Session cost in Codex CLI way too much

Codex Codex CLI
, ,
1

I had a short CLI conversation using GPT-6-Luna medium reasoning in fast mode.

The statusline shows 1.1 Mio Tokens In, 31.2K Tokens Out and $1,82 estimated-threat-cost.

There are different pricings for cache-writes, cache-reads and uncached input. Even if all tokens were expensive cache-writes in fast mode ($0,25/Mio Tokens with GPT-6-Luna), these costs were way too high…

Did anyone experience that these estimated costs are far higher than the actually should be? I am also asking since I noted that the costs displayed in the API console online are way higher than they should be as models were billed which I definitely did not use in this moment (see also separate topic).

2

Firstly, consider logging in with your ChatGPT account subscription if you have one - this presently is a better deal.

Secondly, for Luna, fast mode may be unnecessary - it’s quite a fast model anyway.

3

I do not have the subsription as I only occasionally use it. Fast mode is not neccessary, you are right. I must have activated it by accident. This was just as supporting information.

Key point is that the estimated costs are far beyond the cost as per OpenAI’s official pricing table and I am wondering how Codex CLI comes to this estimation. Probably anyone has seen the same and knows some hints…