Anthropic API Pricing Guide

Anthropic API costs are primarily determined by the Claude model selected and the amount of input and output processed.

How the estimate is calculated

Developers should avoid hard-coding a generic “Claude price” because different Claude models and processing modes have different rates.

For example, Anthropic currently lists Claude Sonnet 4.6 standard global API pricing at $3 per million base input tokens and $15 per million output tokens. Prompt caching and batch processing have their own pricing structures.

What can change the total

A simplified calculation is:

`javascript const inputCost = inputTokens / 1_000_000 * inputRate;

Plan with a realistic usage example

const outputCost = outputTokens / 1_000_000 * outputRate;

const totalCost = inputCost + outputCost; `

Practical checks before you proceed

For a real production cost calculator, you'll likely need more fields.

Useful configuration includes:

Practical checks before you proceed

Model ID Standard input price Output price Cache-write price Cache-read price Batch pricing Effective date

Caching matters particularly for applications repeatedly sending the same large instructions or documents. Batch processing can be useful where jobs do not require interactive latency.

Practical checks before you proceed

Also record actual usage per API request.

This lets engineering teams identify unusually expensive prompts and allows finance teams to understand AI spending by feature or customer.

Practical checks before you proceed

Pricing pages inevitably become outdated as providers release new models. Keep prices in centralized configuration rather than scattering constants throughout your application.

Last reviewed: August 2026. Verify Anthropic's official pricing before using published figures for financial decisions.