Why token counts & costs are 40x?

[lectair]

I send a message to gpt-4-1106-preview which is 6 tokens (in the official OpenAI tokenizer) or 19 characters, and GPT gives me a response of 71 tokens or 273 characters, and below GPT's response it says "gpt-4-1106-preview using 2919 tokens ~= $0. 030610", which is an increase of a x40 of tokens that I can't explain even adding the tokens in my message with the response. Is it a counting bug or am I losing 40 times more money unnecessarily?

Is this happening to anyone else?
Is there any way I can see the raw HTTP requests to the OpenAI endpoint?

Thanks a lot and best regards.

PS: I am using the online version (https://niek.github.io/chatgpt-web)

[Niek]

Are you sure this was the initial message and there was no context sent (i.e. earlier messages in the same thread)?

[lectair]

No, it wasn't the original message. It's because of the settings then, right?

Thanks for the fast response.

[Niek]

You probably have a huge custom system prompt. I checked again and the tokens should be counted correctly. If it's more than what you typed/the context, it's in your system prompt.