genai_pricing

Core package badges:

PYPI

Quality and tooling:

Project/community:

Docs:

Estimate GenAI prompt costs from a unified, auto-updated pricing table. This repo provides a small usage-based cost estimator plus parsers for LiteLLM JSON and markdown pricing tables.

• Parses a local LiteLLM pricing snapshot when present, otherwise falls back to genai_pricing.PRICING_URL
• Computes costs from a usage dictionary with prompt_tokens and completion_tokens
• Includes internal helpers for OpenAI/Gemini-style usage extraction and fallback token counting

Installation

• Python 3.8+
• Packages:
  • tiktoken

pip install tiktoken

Quick start

The included example shows how to estimate cost from an args-like object and token usage dictionary using genai_pricing.estimate_costs. See example.py.

# Minimal example
import os
from openai import OpenAI
from types import SimpleNamespace

from genai_pricing import estimate_costs

"""Estimate the cost of an OpenAI prompt using genai_pricing."""
api_key = os.environ.get("OPENAI_API_KEY")
client = OpenAI(api_key=api_key)
model = "gpt-5.6-sol"
prompt = "Why is the sky blue?"

resp = client.chat.completions.create(
    model=model,
    messages=[{"role": "user", "content": prompt}],
    max_completion_tokens=50,
)

answer = resp.choices[0].message.content
usage = {
    "prompt_tokens": resp.usage.prompt_tokens,
    "completion_tokens": resp.usage.completion_tokens,
}
args = SimpleNamespace(model=model)
estimate = estimate_costs(args, usage)  # <- use this line in your project

print("Cost (USD):", estimate["total_cost"])

Run the example:

python example.py

How cost is computed

Prices are looked up by model name in the pricing table, then applied to token counts:

$$ C = \frac{t_\text{in}}{10^6}. p_\text{in} + \frac{t_\text{out}}{10^6}. p_\text{out} $$

• $t_\text{in}$: prompt tokens
• $t_\text{out}$: completion tokens
• $p_\text{in}$: USD per 1M input tokens
• $p_\text{out}$: USD per 1M output tokens

Provide token counts from your model provider when available. Internal helpers can extract OpenAI- and Gemini-style usage metadata and fall back to tiktoken or a lightweight heuristic when needed.

Pricing table

By default, prices are resolved in this order:

1. model_prices_and_context_window_backup.json in the current working directory
2. model_prices_and_context_window_backup.json near the package/repository location
3. genai_pricing.PRICING_URL, the remote LiteLLM JSON source

The parsed pricing table is cached. Call clear_pricing_cache() after changing the local snapshot or when you want the remote source fetched again.

Testing

The project uses Python’s built-in unittest.

• Run all tests (discovery):

python -m unittest discover -s test -p "*_test.py" -v

API surface

• genai_pricing.estimate_costs
  • Computes a dict with prompt/completion costs and total_cost from an object with a .model attribute and a usage dictionary
• genai_pricing.clear_pricing_cache
  • Clears cached pricing data so the configured source is fetched or read again

Key constant:

• genai_pricing.PRICING_URL — remote table to fetch by default

License

MIT © 2025 Roberto Rossi

Acknowledgements

Pricing data sourced from the AgentOps tokencost table and mirrored (22 Oct 2025) locally at data/pricing_table.md for testing purposes.