modelcost

Calculate LLM API call costs from token usage using price catalogs from multiple sources.

Supported pricing sources:

• litellm (default)
• openrouter
• tokencost

Install

python -m pip install modelcost

CLI

The default command calculates cost, so you can omit the cost subcommand.

# Default (cost)
modelcost gpt-4o 1000 500

# Explicit cost (optional)
modelcost cost gpt-4o 1000 500

# All sources in one run
modelcost --source all gpt-4o 1000 500

# JSON output
modelcost --json gpt-4o 1000 500

List available models:

modelcost models
modelcost models --source openrouter
modelcost models --filter gpt
modelcost models --json

CLI help:

modelcost --help
modelcost models --help

Library

from modelcost.calculator import calculate_cost, list_models

result = calculate_cost("gpt-4o", 1000, 500)

for source in result.available_sources:
    print(f"{source.source}: ${source.total_cost_usd:.6f}")

litellm_cost = next(s for s in result.sources if s.source == "litellm")
print(litellm_cost.price_per_million_input, litellm_cost.price_per_million_output)

models = list_models("openrouter")

Output details

calculate_cost() returns a CostResult with:

• model, input_tokens, output_tokens
• sources: list of SourceCost objects
• available_sources: only sources with prices found

Each SourceCost includes:

• source
• total_cost_usd
• price_per_million_input
• price_per_million_output
• error (when not available)

Caching

openrouter responses are cached in ~/.modelcost_cache.json for 1 hour.

Notes

• Prices are fetched at runtime from the upstream catalogs.
• If a model is missing in a source, that source is marked as unavailable.
• Network sources are fetched in parallel for the all option.