token-budget-pricing

Static per-model pricing lookup table and a ready-made CostModel for token-budget's cost accounting (costModel, maxCost, getUsageReport()).

Pricing changes frequently and this table is a point-in-time snapshot — it will lag reality. Use overrides for up-to-date or custom rates, or supply your own CostModel entirely (e.g. backed by a live pricing API) if that matters for your use case.

Install

npm install @shivam.dixit/token-budget @shivam.dixit/token-budget-pricing

token-budget is a peer dependency (semver range, not pinned).

Usage

import { TokenBudget } from '@shivam.dixit/token-budget';
import { createCostModel } from '@shivam.dixit/token-budget-pricing';

const budget = new TokenBudget({
  maxTokens: 128000,
  model: 'gpt-4o',
  costModel: createCostModel(),
});

budget.addMessage({ role: 'user', content: 'Hello!' });
console.log(budget.stats().cost); // { inputCost, outputCost, totalCost, currency: 'USD' }

Overriding or adding models

const costModel = createCostModel({
  'my-fine-tuned-model': { input: 5e-6, output: 15e-6 }, // USD per token
  'gpt-4o': { input: 3e-6, output: 12e-6 }, // override the built-in rate
});

An unknown model name returns cost 0 rather than throwing — pass overrides for any model not in PRICING_TABLE, or import PRICING_TABLE directly to check what's covered.

What's covered

A snapshot of OpenAI, Anthropic, and Google model pricing — see src/index.ts for the current table. role is accepted by CostModel.costPerToken() for interface compatibility but isn't used here — this model is priced per-model/per-direction, not per-role.

The wider project

Part of the token-budget monorepo — the core package, the other framework/tokenizer adapters, benchmarks, and the flagship coding-agent example all live there. See the compatibility matrix for exactly what every adapter is tested against.

License

MIT