claude-pricing

Calculate and track Claude API costs from token usage. Bundled pricing table + custom overrides. Zero dependencies.

import { calculateCost, formatUSD, CostTracker } from "claude-pricing";

// One-shot cost calculation
const cost = calculateCost("claude-opus-4-7", {
  input_tokens: 1500,
  output_tokens: 800,
  cache_read_input_tokens: 5000,
});
console.log(formatUSD(cost.total));  // "$0.0900"

// Session tracker with budget alert
const tracker = new CostTracker({
  budgetUSD: 1.0,
  onBudgetExceeded: (total, budget) =>
    alert(`Budget blown: ${formatUSD(total)} > ${formatUSD(budget)}`),
});

tracker.add("claude-opus-4-7", usage, { label: "planning" });
tracker.add("claude-sonnet-4-6", usage, { label: "draft" });

console.log(formatUSD(tracker.total().total));
console.log(tracker.byModel());

Why this exists

You're running Claude API in production. You want to know:

• What did this one call cost? The SDK returns usage but doesn't tell you the dollars.
• Did this session blow my budget? You want a running total across an agent loop.
• Where's my money going? Per-model breakdown for cost optimization.
• What about cache and batch? The pricing math is non-obvious (1.25× input for cache write, 50% off for batch).

This package is ~180 lines of TypeScript. Zero deps. Bundled with up-to-date Anthropic pricing, but everything is overridable because list prices change and you may have a custom contract.

Install

npm install claude-pricing

API

calculateCost(model, usage, options?)

function calculateCost(
  model: string,
  usage: TokenUsage,
  options?: { pricing?: ModelPricing | Record<string, ModelPricing>; batch?: boolean }
): CostBreakdown;

Returns { input, output, cacheWrite, cacheRead, total } — all in USD.

CostTracker

const tracker = new CostTracker({
  pricing?: customTable,
  budgetUSD?: 5.00,
  onBudgetExceeded?: (total, budget) => ...,
});

tracker.add(model, usage, { label?: 'step-1', batch?: true });
tracker.total();      // { input, output, cacheWrite, cacheRead, total }
tracker.byModel();    // { 'claude-opus-4-7': { calls, usage, breakdown }, ... }
tracker.reset();
tracker.getEntries(); // readonly array of { model, usage, breakdown, timestamp, label }

formatUSD(amount)

Smart-precision formatter:

• $1.23 for amounts ≥ $1
• $0.0123 for amounts ≥ $0.01
• $0.000123 for sub-cent

ANTHROPIC_PRICING

The bundled default table (USD per million tokens):

| Model | Input | Output | Cache write | Cache read | |---|---|---|---|---| | claude-opus-4-7 | $15 | $75 | $18.75 | $1.50 | | claude-sonnet-4-6 | $3 | $15 | $3.75 | $0.30 | | claude-haiku-4-5 | $0.80 | $4 | $1.00 | $0.08 |

⚠️ Snapshot taken 2026-05. Verify against Anthropic pricing for production. Pass your own table via options.pricing for current rates.

Recipes

Override pricing for current rates

const myPricing = {
  "claude-opus-4-7": { input: 14, output: 70 },  // discounted contract
};
const cost = calculateCost(model, usage, { pricing: myPricing });

Use for any custom/fine-tuned model

const cost = calculateCost(
  "my-fine-tune",
  usage,
  { pricing: { input: 8, output: 24 } }  // single ModelPricing
);

Pair with claude-stream-collector

import { collectStream } from "claude-stream-collector";
import { calculateCost, formatUSD } from "claude-pricing";

const result = await collectStream(stream);
const cost = calculateCost(result.model, result.usage);
console.log(formatUSD(cost.total));

The TokenUsage shape is compatible — pass result.usage directly.

Track an agent loop with budget cap

const tracker = new CostTracker({
  budgetUSD: 0.10,
  onBudgetExceeded: () => { throw new Error("Cost limit reached"); }
});

for (const step of agent.run()) {
  tracker.add(step.model, step.usage, { label: step.name });
  // throws automatically when total > $0.10
}

console.log(tracker.byModel());

Batch API (50% off)

calculateCost(model, usage, { batch: true });
// Or per add():
tracker.add(model, usage, { batch: true });

Tests & build

npm install
npm test       # vitest run
npm run build  # tsc

License

MIT — © 2026 xiangnuans

Part of a series

The 3rd package in small, focused Claude SDK helpers built in public.

• ✅ claude-stream-collector — Consume streaming events into typed result
• ✅ claude-retry — Smart retry with backoff
• ✅ claude-pricing — Cost calculation + session tracker (this package)
• 🚧 claude-prompt-toolkit — Prompt templating, versioning

Follow the journey: ship-log-2026.