ai-api-pricing

Zero-runtime-dependency TypeScript/JavaScript access to the source-linked AICostBudget AI API Pricing Dataset.

Data, history and issue tracker: aicostbudget/ai-api-pricing-data on GitHub. Spot a wrong price? Open an issue with the official source link.

npm install ai-api-pricing

import { estimateCost } from "ai-api-pricing";
const estimate = await estimateCost({
  modelId: "gpt-4.1", inputTokens: 1_000_000, outputTokens: 100_000,
});
console.log(estimate.total, estimate.officialSourceUrl, estimate.lastVerifiedAt);

The package exports loadPrices, getPrice, listModels, estimateCost, and compareModels. Live loading uses the GitHub Pages JSON with a 10-second timeout and falls back to the bundled snapshot on any fetch or validation error. Check the returned source field to distinguish live from bundled data. Missing required prices raise a typed MissingPriceError; the package never substitutes another rate.

From price per token to cost per result

A price estimate is the cost of one call. To measure what one correct result costs on your models, including retries and reasoning tokens, use the open LLM cost-per-task benchmark: 16 tasks with code-only validators, run with your own API keys, priced from this dataset.

Refresh the bundled snapshot

From the repository root, run:

npm run refresh-bundle --workspace ai-api-pricing

The generator reads the checked-in api/v1/prices.json, embeds its generated_at value, and rewrites src/generated/bundled.ts. Review and commit that generated file with the package change.

Data license and citation

Code is MIT-licensed. Pricing data is licensed under CC BY 4.0. Attribution: AICostBudget AI API Pricing Dataset, https://aicostbudget.com/en/datasets/ai-api-pricing, DOI 10.5281/zenodo.23087250.

Prices are verified against official provider pages; always confirm before billing decisions.