TokenCost — LLM Price Calculator for 400+ Models

What it is

TokenCost is a Python library that counts tokens and estimates USD costs for over 400 LLM models. In three lines of code, you can calculate the cost of any prompt or completion across OpenAI, Anthropic, Mistral, AWS Bedrock, and other providers. The library runs entirely client-side, using local tokenizers to count tokens without sending data to any API.

Developers building LLM applications who need to track costs, set budgets, or compare pricing across providers benefit from TokenCost. It is especially useful for applications that route between multiple models and need real-time cost visibility.

How it saves time or tokens

TokenCost eliminates the need to manually look up pricing pages and count tokens for each provider. Instead of maintaining a spreadsheet of model prices, you call one function and get the USD cost. This saves time during development and enables automated cost monitoring in production. Pre-checking prompt costs before sending them to the API prevents budget overruns.

How to use

1. Install TokenCost via pip
2. Import the cost calculation functions
3. Pass your prompt text and model name to get the USD cost

Example

from tokencost import calculate_prompt_cost, calculate_completion_cost

# Calculate prompt cost
prompt = 'Explain quantum computing in simple terms.'
cost = calculate_prompt_cost(prompt, model='gpt-4o')
print(f'Prompt cost: ${cost:.6f}')

# Calculate completion cost
completion = 'Quantum computing uses qubits...'
cost = calculate_completion_cost(completion, model='gpt-4o')
print(f'Completion cost: ${cost:.6f}')

# Compare across models
for model in ['gpt-4o', 'claude-sonnet-4-20250514', 'mistral-large']:
    c = calculate_prompt_cost(prompt, model=model)
    print(f'{model}: ${c:.6f}')

Related on TokRepo

• AI tools for monitoring — Browse cost monitoring and observability tools
• AI gateway tools — Explore API gateway solutions with built-in cost tracking

Common pitfalls

• Model prices change frequently; update TokenCost regularly to keep the pricing database current
• Token counts are estimates based on local tokenizers; actual API billing may differ slightly due to special tokens
• Some models use different tokenizers; ensure TokenCost supports your specific model variant for accurate counts