Below is a concise explanation with examples of how to calculate token usage (and therefore cost) for an API request:

How Tokens Are Counted

When you send a chat completion API call, the total tokens include:

• System messages (if provided)
• User messages
• Assistant messages (the output)

All tokens—both those sent (prompt tokens) and those received (completion tokens)—are summed to determine the final count. For example, if the API response shows:

JSONCopied
{
  "usage": {
    "prompt_tokens": 19,
    "completion_tokens": 10,
    "total_tokens": 29
  }
}

Then you are billed for 29 tokens, which includes tokens from the system, user, and assistant messages.

How to Estimate Tokens for a Request

1. Use the API response:
   After making a call, check the usage field. For example:

   PYTHONCopied
   total_tokens = response['usage']['total_tokens']

   This count includes every token processed as part of the request.

2. Estimate tokens before making a call using the tiktoken library:
   You can run a local estimate to see how many tokens your messages might use, including system messages:

   PYTHONCopied
   import tiktoken
   
   # Use the appropriate encoding (often "cl100k_base" for gpt-3.5-turbo).
   encoding = tiktoken.get_encoding("cl100k_base")
   
   # Define your messages (system, user, etc.)
   messages = [
       {"role": "system", "content": "You are an expert in American cuisine, and creating different dishes from items found in US grocery stores."},
       {"role": "user", "content": """I will give you the name of a common product found in a US grocery store.  

I would like you to tell me the dish most commonly consumed in the US that contains this item as its featured ingredient.
If possible, ensure the dish is one that most would consider to be 'American cuisine.'
Please restrict your answer to one item, and include only the name of the dish.

The item is:

Boneless skinless chicken thighs"""}
]

Count tokens for each message

total_tokens = 0
for message in messages:
message_tokens = len(encoding.encode(message["content"]))
total_tokens += message_tokens
print(f"Estimated tokens for prompt: {total_tokens}")

*Note:* When you later get the assistant’s reply, its tokens (completion tokens) will add to this cost. Always remember to include system messages in your estimation since they count toward your billed tokens.

### Cost Calculation

Once you have the total token count, you can estimate the cost by multiplying with your model’s price per token. Check the [OpenAI Pricing page](https://openai.com/api/pricing) for the specific cost details of your model, since the price may vary between input and output tokens.

### Summary

- **Yes**, the system message affects token usage.
- **Calculate cost** by:
1. Summing tokens from all messages (system, user) for the prompt.
2. Adding the tokens from the assistant’s reply.
3. Using the API response’s `usage` field or a library like tiktoken to estimate before making calls.
- Multiply the total tokens by the cost per token (reference the pricing page) to determine the final cost.

This approach will help you estimate and manage the costs for your web app accurately.