Before you build a Claude-powered automation, you need to know what it’s going to cost. Too many people set up a scenario, turn it on, and then find out their bill three weeks later.

This post walks through exactly how to calculate your Claude API bill in advance, with worked examples for the most common automation use cases. No spreadsheet required — just multiplication.

The Basic Formula

Claude API pricing is based on tokens, not requests. A token is roughly 3/4 of a word (English), so 1,000 tokens ≈ 750 words.

Every API call has two token counts:

Output costs roughly 5x more than input. Track both.

2026 Claude Pricing

Per million tokens, as of April 2026:

Model Input Output
Claude Haiku 4.5 $1 $5
Claude Sonnet 4.6 $3 $15
Claude Opus 4.7 $15 $75

(Always verify current pricing at anthropic.com/pricing before making decisions.)

Convert to per-thousand-tokens (easier for small calculations):

Model Input per 1k Output per 1k
Haiku 4.5 $0.001 $0.005
Sonnet 4.6 $0.003 $0.015
Opus 4.7 $0.015 $0.075

The Calculation

For any given workflow:

Related Free Resource

Ready to build your first one?

The free QuickStart walks you through a complete working Make.com + Claude scenario in about 30 minutes. No email required, name your price (including $0).

Download Free → See all courses 
Cost per run = (input_tokens × input_rate) + (output_tokens × output_rate)

Monthly cost = Cost per run × runs per month

That’s it. All the math below is just filling in numbers.

Worked Example 1: Email Summariser

Workflow: every incoming email gets a one-sentence summary added to a spreadsheet.

Per-call token counts:

Model choice: Haiku 4.5 (simple task, high volume)

Cost per email:

At 500 emails/month: $0.625

At 2,000 emails/month: $2.50

At 10,000 emails/month: $12.50

Realistically — this workflow is free.

Worked Example 2: Customer Reply Generator

Workflow: draft personalised replies to customer enquiries. Higher stakes than summarisation.

Per-call token counts:

Model choice: Sonnet 4.6 (writing quality matters)

Cost per reply:

At 100 replies/month: $1.11

At 1,000 replies/month: $11.10

At 10,000 replies/month: $111

Still very reasonable for the value.

Worked Example 3: Document Analyser

Workflow: analyse long business documents and extract key findings.

Per-call token counts:

Model choice: Sonnet 4.6 or Opus 4.7 depending on complexity

Cost per document (Sonnet):

Cost per document (Opus):

At 100 documents/month on Sonnet: $14

At 100 documents/month on Opus: $72

The 5x cost difference is the reason to start with Sonnet and only upgrade if output quality demands it.

Worked Example 4: Always-On Chatbot

Workflow: website chatbot handling customer conversations.

Per-conversation token counts:

Average per-message cost (Sonnet):

By message 8, you’re sending: system prompt + 7 previous messages (growing context):

Cost per full conversation:

At 100 conversations/day: $7.20/day = $216/month At 20 conversations/day: $1.44/day = $43/month At 5 conversations/day (small business): $0.36/day = $11/month

Token Counting in Practice

You don’t need to manually count tokens. Tools:

  1. Anthropic’s Usage page — shows actual token usage per API call, retroactively
  2. Anthropic tokenizer — paste text, see token count
  3. Rough estimate — word count × 1.3 = approximate token count

For planning, rough estimates are fine. For optimising, check actual usage after running the workflow for a week.

Optimisation Patterns That Cut Bills

Three patterns cut costs 50-80% without hurting quality:

1. Prompt Caching (90% savings on repeated context)

If you have a long system prompt used across many calls, Anthropic’s prompt caching lets subsequent calls use the cached version at 10% of normal input cost.

Example: a 2,000-token system prompt used 1,000 times a month:

90% savings on the system prompt portion.

2. Model Routing

First module: Haiku classifies. Only the cases that matter go to Sonnet for full processing.

Example: 500 emails/day, 60% are spam:

50% savings from filtering out obvious non-work.

3. Output Caps

Setting max_tokens on every module prevents runaway generation. A scenario without caps that accidentally generates 4,000 tokens instead of 400 = 10x the cost of that call.

Set sane caps: - Classification: 100-200 - Short reply: 300-500 - Long reply: 800-1500 - Report: 2000-3000

Setting a Hard Monthly Limit

The final safety net: Anthropic lets you set a monthly spend limit. Go to Billing → Plans → Usage limits. Set this to whatever you’re comfortable with.

If the workflow goes haywire and tries to spend beyond your limit, Anthropic pauses your API access. You get notified. No $2,000 surprise on a card you forgot was connected.

Do this. Always. Even if your budget is generous.

My Rule for New Projects

Before turning on any new automation, I calculate the expected monthly cost and multiply by 2x. That’s my budget. If the 2x number is uncomfortable, I redesign the workflow (cheaper model, fewer calls, shorter outputs) before deploying.

Costs only surprise you if you don’t plan. Do the math first; deploy second.

Next Steps

The Implementation Blueprint course ($29) includes a spreadsheet template for estimating costs across multi-step workflows — particularly useful when you’re using Haiku, Sonnet, and Perplexity together and need to model total spend.

For simpler starter scenarios, the free Quick Start uses Haiku exclusively, so you can get hands-on without worrying about costs.

Last updated: 20 April 2026. Pricing changes — always verify current numbers before building cost models.

Related Free Resource

Ready to build your first one?

The free QuickStart walks you through a complete working Make.com + Claude scenario in about 30 minutes. No email required, name your price (including $0).

Download Free → See all courses