TOKENSE.

TOKENSE / Blog / How Claude API Pricing Works: Calculate Your Real Cost Per Message

料金・計算

How Claude API Pricing Works: Calculate Your Real Cost Per Message

Published: 2026-07-29 ・ TOKENSE Blog

Claude API isn't a flat monthly fee — it's metered, pay-per-token pricing. Once you understand the formula, you can estimate exactly what a conversation will cost. Here's the official Anthropic pricing table and worked examples.
In this article
  1. How Claude API billing works
  2. Official pricing (as of July 2026)
  3. Calculating the cost of one message
  4. Cutting costs with caching and batching
  5. FAQ

1. How Claude API billing works

Unlike a flat-fee product like ChatGPT Plus, the Claude API (Anthropic's direct model access) bills per token — the smallest unit of text a model processes. As a rough guide, 1 token is about 4 English characters, or roughly 0.75 words.

Pricing is split between input (what you send) and output (what the model generates), and output tokens are always priced higher. Longer answers cost more.

2. Official pricing (as of July 2026)

Per Anthropic's official documentation (verified 2026-07-29 at platform.claude.com/docs/en/about-claude/pricing), prices per million tokens (MTok) are:

ModelInput (1M tok)Output (1M tok)Typical use
Claude Haiku 4.5$1.00$5.00Light questions, summaries, fast tasks
Claude Sonnet 5
(through Aug 31, 2026)
$2.00$10.00General-purpose, balanced tasks
Claude Opus 5$5.00$25.00Complex reasoning, highest quality

Source: Anthropic official Pricing docs (retrieved 2026-07-29). Sonnet 5 moves to $3/$15 per MTok starting September 1, 2026, per the same page.

3. Calculating the cost of one message

FormulaCost = (input tokens ÷ 1,000,000 × input price) + (output tokens ÷ 1,000,000 × output price)

Example: a short ~500-token question answered in ~800 tokens by Haiku 4.5:

Run the same exchange on Opus 5, and the cost rises roughly 5x, since both input and output prices are 5x higher. Model choice is the single biggest lever on your bill.

Manual tracking doesn't scaleIn real conversations, prior turns are resent as context each time, so input tokens grow as a chat gets longer. That's why TOKENSE shows the real cost of every message automatically, plus running daily/monthly totals.

4. Cutting costs with caching and batching

TOKENSE automatically compresses long chat history and leans on caching, so costs stay flat even as conversations grow — plus an "Audit" feature to check whether a cheaper model was good enough for your specific question.

FAQ

Q. Is the Claude API free?
A. New Anthropic accounts sometimes get a small free credit, but usage beyond that is billed per token. TOKENSE itself is free — you only pay Anthropic for API usage.
Q. How much does a typical question cost?
A. It depends on the model and answer length, but a short question on Haiku 4.5 usually costs well under a cent. TOKENSE shows the exact figure for every message.
Q. Will pricing change?
A. Yes — per Anthropic's own page, Sonnet 5 moves from $2/$10 to $3/$15 per million tokens on September 1, 2026. Always check the official page for the latest rates.

TOKENSE is free, no sign-up required

Paste your own Anthropic API key and start chatting with cost telemetry built in. No account, no server.

Related reading

Is ChatGPT Plus Too Expensive? A Pay-Per-Use API Alternative TOKENSE: A BYOK Claude Client with Automatic Cost Tracking