A simple light weight wrapper for OpenAI's API that tracks the cost of each request. It also provides simulation mode to test the cost of a request without actually sending it to OpenAI.

pip install openai-cost-tracker

Import the query_openai function from openai_cost_tracker:

from openai_cost_tracker import query_openai

Use query_openai function as a drop-in replacement for openAI's completion function, such as openai.Completion.create(). The query_openai will return the same response as the original function, but will also print the cost of the request.

Turn on simulation to test the cost of a request without actually sending it to OpenAI. Turn on print_cost to print the cost of each request.

import openai 
from openai_cost_tracker import query_openai

openai.api_key = "<YOUR_OPENAI_API_KEY_HERE>"

prompt = "Hello World!"  # your prompt here

response = query_openai(
    model="gpt-4-0125-preview",  # support gpt-4-0125-preview,  gpt-3.5-turbo-1106,  gpt-4
    messages=[{'role': 'user', 'content': prompt}],            
    max_tokens=5,
    # rest of your OpenAI params here ...
    simulation=True,  # set to True to test the cost of a request without actually sending it to OpenAI 
    print_cost=True   # set to True to print the cost of each request
)     

print(response["choices"][0]["message"]["content"])

Input tokens: 10 | Output tokens: 5 | Cost: $0.0003 | Total: $0.0003
Hello! How can I

• Under the simulation model, the cost of request is calculated based on given max_tokens and number of input tokens using the OpenAI tokenization scheme. This is an approximation of the upper bond of the actual cost.
• The code is mainly implemented for GPT-3.5 and GPT-4 models, but it will work for other openAI models by customizing the utils.py and estimator.py files. Feel free to open an issue or PR if you need help with that.
• The implementation is based on this repo.