I tried calculating this twice.
First time I got $0.30 cents per minute, second time I got $0.024.
Could someone be so kinda and help me out? I don’t want to get this wrong. I know openai said they cut down the costs by 80% recently, but I want to double check with the community, or anyone available and down to share their calculations
Pricing :
https://platform.openai.com/docs/pricing
It is shown as a price per token, not per minute (like Whisper is).
How is an audio token calculated ? is it the same “token” as text ?
I quickly looked fo it, but didn’t see anything about it on main pricing page, or on the model’s page.
Edit:
I just saw that your post date from December 2024. There was a new release on December 17, lowering the price from 100$/1M tokens to 40$/1M. (for inputs). Output got from 200$ to 80$ /1M tokens.
Audio is tokenized by a proprietary method.
Stated on the old pricing page, referring to $100/1m in and $200/1m out:
***Audio input costs approximately 6¢ per minute; Audio output costs approximately 24¢ per minute
Then, in a “chat”, every bit of past input and generation becomes another input cost upon every generation triggering, giving growing repeated costs in a longer conversational exchange.
Normal spoken audio conveys the text spoken about 5-10x less efficiently in tokens than a transcription, beyond the higher cost of audio.