Generative AI For The Rest Of US

How Costs are Calculated

AI services like ChatGPT, Google Gemini, Claude, and similar platforms typically charge users based on the same principle. First, the cost is calculated based on whether you're using the service as a regular user (via the web interface) or as a developer (via the API). For regular users, the cost is usually a fixed monthly fee, while the cost for developers using the API is calculated based on the number of tokens used. Before diving into the specifics of each platform, let's understand what a token is and how it affects the cost.

A token is a unit of computation that represents a word or a subword in the input text (prompt) or the output text generated by the model. Each provider (such as OpenAI, Google, Claude, etc.) defines the size of a token based on the model architecture and the specific requirements of the service. So, there isn't a fixed definition of a token that applies across all platforms; however, it tends to be around 4 characters long. For OpenAI's GPT models, as an example, a token is typically 4 characters long, so 100 tokens would be around 75 words. This also means that a common sentence of about 15 to 20 words would use approximately 20 to 27 tokens. You can play with the online OpenAI tokenizer to get a better understanding of how tokens are calculated. Let's see an example:

Hello, how are you doing today?

In the above sentence, there are 8 tokens. This is the list:

- [Hello]
- [,]
- [ how]
- [ are]
- [ you]
- [ doing]
- [ today]
- [?]

Each model has a different cost per token, which is used to calculate the total cost of using the service. Also, there are 3 factors you should take into account when you calculate the cost of using an AI service:

1. The number of tokens in the input prompt: The text you provide to the model.