llm-costs

LLM cost calculator for major providers. Calculates API costs from token usage data.

Installation

pip install llm-costs

Or with uv:

uv add llm-costs

Usage

from llm_costs import calculate_cost, get_model_pricing

# Calculate cost from LangChain UsageMetadata structure
result = calculate_cost(
    provider="anthropic",
    model="claude-sonnet-4-20250514",
    usage={
        "input_tokens": 1000,
        "output_tokens": 500,
        "total_tokens": 1500,
    },
)

print(f"Cost: ${result['cost']:.6f}")  # Cost: $0.010500
print(f"Input: ${result['breakdown']['input_cost']:.6f}")
print(f"Output: ${result['breakdown']['output_cost']:.6f}")

With Prompt Caching

result = calculate_cost(
    provider="anthropic",
    model="claude-sonnet-4-20250514",
    usage={
        "input_tokens": 1000,
        "output_tokens": 500,
        "total_tokens": 1500,
        "input_token_details": {
            "cache_read": 5000,
            "cache_creation": 0,
        },
    },
)

Batch Pricing

result = calculate_cost(
    provider="anthropic",
    model="claude-sonnet-4-20250514",
    usage={"input_tokens": 1000, "output_tokens": 500, "total_tokens": 1500},
    batch=True,  # 50% discount
)

Get Model Pricing Info

pricing = get_model_pricing("anthropic", "claude-sonnet-4-20250514")
print(f"Input: ${pricing['input']}/MTok")   # Input: $3.0/MTok
print(f"Output: ${pricing['output']}/MTok") # Output: $15.0/MTok

List Available Models

from llm_costs.calculator import list_models, list_providers

providers = list_providers()  # ['anthropic', 'openai', 'google']
models = list_models("anthropic")  # ['claude-opus-4-5-20251101', ...]

Supported Providers

Anthropic

Prices per million tokens. Long context pricing (>200K tokens) applies to Sonnet models.

OpenAI

Prices per million tokens (Standard tier).

Google

Prices per million tokens (Paid tier). Long context pricing (>200K tokens) applies to Pro models.

Usage Schema

The library accepts token usage in LangChain's UsageMetadata format:

{
    "input_tokens": int,
    "output_tokens": int,
    "total_tokens": int,
    "input_token_details": {
        "cache_read": int,      # Cached tokens read
        "cache_creation": int,  # Tokens written to cache
    },
    "output_token_details": {
        "reasoning": int,       # Reasoning tokens (o1/thinking models)
    },
}

Return Value

calculate_cost() returns a CostResult:

{
    "cost": 0.0105,           # Total cost in USD
    "currency": "USD",
    "breakdown": {
        "input_cost": 0.003,
        "output_cost": 0.0075,
        "cache_read_cost": 0.0,      # If applicable
        "cache_creation_cost": 0.0,  # If applicable
    },
    "pricing_used": {
        "input_per_mtok": 3.0,
        "output_per_mtok": 15.0,
        "batch_applied": False,
        "long_context_applied": False,
    },
}

License

MIT