Note While GPT4-Turbo has already been released, and it is priced lower and is also faster, the Markdown output from GPT4-Turbo is not very consistent at the moment. I expect this to improve when it comes out of preview, but for now the Markdown generated by GPT4-Turbo is not consistent enough for me to parse the output during the appropriate step in the workflow
As of this writing, the cost for using GPT-4 is 3 cents per 1K prompt tokens, and 6 cents for 1K sampled tokens.
Please update your estimates based on the cost when you are reading this.
Prompt vs sampled tokens
The input we send to GPT4 is used for counting prompt tokens, and the output that GPT4 generates is used for counting sampled tokens.
As you can see, the price of sampled tokens is double the price of prompt tokens.
Estimating number of tokens per minute
There are a few online tools which can help you with these estimates.
I will use the gpt-tokenizer playground tool.
Let us consider the nearly 6 minute long video here
As you can see, this will produce approximately 1200 prompt tokens and 1200 sampled tokens (because the input and output length are nearly identical) for both the formatting step and the paragraph subheading generation step.
In other words, we consume about 200 prompt tokens and 200 sampled tokens for each audio minute for two different GPT4 requests. This equates to 400 prompt tokens and 400 sampled tokens per minute for the complete workflow.
400 prompt tokens = 0.4 * 3c = 1.2 cents per minute
400 sampled tokens = 0.4 * 6c = 2.4 cents per minute
Total = 3.6 cents per minute
Professional transcription used to cost around $2 per minute for producing a well formatted video transcript (and minus the subheadings), so you can see that this is a dramatic reduction in cost, even though there is certainly a small reduction in quality.
Please remember these are only estimates and will depend on the speed at which you speak in your videos.
About this website BotFlo1 was created by Aravind Mohanoor as a website which provided training and tools for non-programmers who were2 building Dialogflow chatbots. This website has now expanded into other topics in Natural Language Processing, including the recent Large Language Models (GPT etc.) with a special focus on helping non-programmers identify and use the right tool for their specific NLP task. For example, when not to use GPT 1 BotFlo was previously called MiningBusinessData. That is why you see that name in many videos 2 And still are building Dialogflow chatbots. Dialogflow ES first evolved into Dialogflow CX, and Dialogflow CX itself evolved to add Generative AI features in mid-2023