tokencost-js

A JavaScript/TypeScript port of the tokencost library that works in any JavaScript runtime.

Demo that estimates the price across multiple LLMs: llmcomp.backmesh.com

This library calculates token counts and costs for prompts and completions of various LLM models. It supports Anthropic Claude models using their token counting API endpoint from their official SDK. This requires for the Anthropic SDK to be configured via environment variables for Anthropic cost calculations to work. If you calculate the costs of a completion for an Anthropic model, it will fail unless you set these environment variables.

Caution

Make sure you do not expose your ANTHROPIC_API_KEY on the internet! You should never expose any secrets in the bundle of a web or mobile app. The demo site supports Anthropic calculations because it uses a Backmesh LLM API Gatekeeper for communication with the Anthropic API. The ANTHROPIC_BASE_URL is set to the Backmesh URL and the ANTHROPIC_API_KEY is set to a JWT generated using anonymous Firebase authentication. Backmesh is an open source Backend as a Service (BaaS) available here.

Installation

npm install tokencost-js

Usage

Basic Usage

import { 
  countMessageTokens, 
  countStringTokens, 
  calculatePromptCost, 
  calculateCompletionCost, 
  calculateAllCostsAndTokens 
} from 'tokencost-js';

async function main() {
  // Example 1: Count tokens in a string
  const text = "Hello, world! This is a test message to count tokens.";
  const model = "gpt-4";
  const stringTokens = await countStringTokens(text, model);
  console.log(`Example 1: String "${text}" has ${stringTokens} tokens with model ${model}`);

  // Example 2: Count tokens in messages
  const messages = [
    { role: "system", content: "You are a helpful assistant." },
    { role: "user", content: "What is the capital of France?" },
    { role: "assistant", content: "The capital of France is Paris." },
    { role: "user", content: "What about Germany?" }
  ];
  const messageTokens = await countMessageTokens(messages, "gpt-3.5-turbo");
  console.log(`Example 2: Messages have ${messageTokens} tokens with model gpt-3.5-turbo`);

  // Example 3: Calculate prompt cost
  const promptCost = await calculatePromptCost("What is the capital of France?", "gpt-4");
  console.log(`Example 3: Prompt cost: $${promptCost.toFixed(6)} with model gpt-4`);

  // Example 4: Calculate completion cost
  const completionCost = await calculateCompletionCost("The capital of France is Paris.", "gpt-4");
  console.log(`Example 4: Completion cost: $${completionCost.toFixed(6)} with model gpt-4`);

  // Example 5: Calculate all costs and tokens
  const allCosts = await calculateAllCostsAndTokens(
    "What is the capital of France?",
    "The capital of France is Paris.",
    "gpt-4"
  );
  console.log("Example 5: All costs and tokens:");
  console.log(`  Prompt tokens: ${allCosts.promptTokens}`);
  console.log(`  Prompt cost: $${allCosts.promptCost.toFixed(6)}`);
  console.log(`  Completion tokens: ${allCosts.completionTokens}`);
  console.log(`  Completion cost: $${allCosts.completionCost.toFixed(6)}`);
  console.log(`  Total cost: $${(allCosts.promptCost + allCosts.completionCost).toFixed(6)}`);
}

main().catch(console.error); 

Working with Messages

import { calculatePromptCost, type Message } from 'tokencost-js';

const messages: Message[] = [
  { role: "system", content: "You are a helpful assistant." },
  { role: "user", content: "Tell me about the solar system." },
  { role: "assistant", content: "The solar system consists of the Sun and everything that orbits around it." },
  { role: "user", content: "How many planets are there?" }
];

const cost = await calculatePromptCost(messages, "gpt-4");
console.log(`Cost to process these messages: $${cost.toFixed(6)}`);

Updating Token Costs

The library comes with a static set of token costs, but you can update them with the latest costs from the LiteLLM cost tracker:

import { updateTokenCosts, TOKEN_COSTS } from 'tokencost-js';

// Update token costs
await updateTokenCosts();
console.log("Token costs updated:", TOKEN_COSTS);

Supported Models

The library supports a wide range of models including:

• OpenAI models (GPT-3.5, GPT-4, GPT-4o, etc.)
• Claude models (with estimated token counts)
• And many more