Re:OpenAI Tokenizer
by engrefoobiz on Wednesday April 05, 2023 @11:26AM
(#35455801)
GPT-4 costs 30 times more than gpt-3.5-turbo and 60ktimes more if you use the 32k token gpt-4 model. It's by far their most expensive service!
I'm using gpt-3.5-turbo, also for coding, and honestly it does just fine.
Re:OpenAI Tokenizer
by SubiculumCode on Wednesday April 05, 2023 @11:22AM
(#35455726)
excuse my ignorance, but I thought it was $20 per month.
Re:OpenAI Tokenizer
by bikingbismuth on Wednesday April 05, 2023 @11:23AM
(#35455742)
That is for the interactive chat experience. API calls are sold a la carte. Details here https://openai.com/pricing.
Re:OpenAI Tokenizer
by SubiculumCode on Wednesday April 05, 2023 @11:43AM
(#35456050)
thanks!
Re:OpenAI Tokenizer
by MuffinFlavored on Wednesday April 05, 2023 @02:27PM
(#35458473)
It would be cool if they told their $20/mo users "here's how much your past 30 day usage would have cost if we billed you via the API (aka how many tokens/sessions/chats/whatever did you use)
The seems remarkably cheap compared to engineering hours.
To do....?
Re:OpenAI Tokenizer
by smallerfish on Wednesday April 05, 2023 @12:48PM
(#35457000)
3.5-turbo can definitely understand code, just like ChatGPT can. GPT4 is better at complex tasks, but 3.5-turbo is always worth evaluating.
~100 lines of code + whitespace
1300-1900 tokens
So if I fed this to OpenAI and said "how can I make this file better/improve upon it", it would have cost:
between $0.03 and $0.12 for this one file using GPT-4
not sure I could use gpt-3.5-turbo since it says it is for chat and not code?
Does that sound right? $0.05 for every file of source code scanned sounds too high for realistic usage. Even $0.01 sounds high? Modern company might have 1,000,000+ files of code, no?