You can read more details about it here in this post:
And here:
https://openai.com/codex/
I use Codex Extension with 5.4 model, for local project. I added 1000 credits, thought it is good since it claims 1000 credits are estimated to be equal to 250-1200 messages. However, each simple coding message (with closed IDE context) consumed more than 10 credits. it makes less than 100 messages. It is not nice stealing this way, sucking blood of customers. I urge you to re-consider the cost calculation.