“How much does the Claude API actually cost?” is the most common question I get — usually right after someone has set up a Make.com scenario and is afraid their card is about to get hit hard.
Short answer: if you’re sensible, very little. If you’re not, a lot. This post breaks down exactly how Anthropic charges for Claude API usage in 2026, what real-world workflows cost, and where people accidentally burn through credit.
I’ll use numbers from my own usage wherever I can, not speculation.
How Claude API Pricing Actually Works
Anthropic charges per token, not per request. A token is roughly 3/4 of a word — so 1,000 tokens is about 750 English words.
Every API call has two token types that cost different amounts:
- Input tokens — everything you send to Claude: your system prompt, the user message, any attached conversation history
- Output tokens — what Claude generates back
Output tokens cost roughly 5 times more than input tokens. This is across every Claude model, and it’s why response length matters so much to your bill.
2026 Claude Model Pricing
As of April 2026, Anthropic’s public pricing (per million tokens):
| Model | Input | Output | Best For |
|---|---|---|---|
| Claude Haiku 4.5 | ~$1 | ~$5 | High-volume automation, quick tasks |
| Claude Sonnet 4.6 | ~$3 | ~$15 | Balanced reasoning and cost |
| Claude Opus 4.7 | ~$15 | ~$75 | Complex analysis, research |
Important caveat: These prices change. Anthropic has raised and lowered prices multiple times across the 4.x generation. Always cross-check against anthropic.com/pricing before building cost models.
What Actual Workflows Cost
Here are real examples from the Make.com scenarios I run.
All 3 courses + AI Playbook — $49
Everything: QuickStart, Implementation Blueprint, and the AI Automation Playbook (reference PDF with prompt templates, cost calculator, and multi-API routing patterns). One payment, lifetime access.
Example 1: Email Summarisation
- Input: 800-token email + 200-token system prompt = 1,000 tokens
- Output: ~100 tokens (one-sentence summary)
- Model: Haiku 4.5
- Cost per run: $0.001 + $0.0005 = $0.0015
If you run this on 500 emails a month, total cost is about $0.75. Genuinely. That’s less than a coffee.
Example 2: Customer Chat Response
- Input: 1,500-token conversation history + 300-token system prompt
- Output: ~200 tokens
- Model: Sonnet 4.6
- Cost per run: $0.0054 + $0.003 = $0.0084
At 100 customer messages a day (3,000/month), this costs about $25/month.
Example 3: Long Document Analysis
- Input: 50,000-token PDF + 500-token prompt
- Output: ~2,000 tokens structured summary
- Model: Opus 4.7
- Cost per run: $0.758 + $0.15 = $0.91
At 50 documents/month, this is $45. Worth it for specialist analysis, not worth it if Sonnet would have been adequate.
Where Costs Go Wrong (From Mistakes I’ve Made)
These are the patterns that turn a $10/month workflow into a $500/month surprise:
1. Using Opus for everything
Opus is incredible for reasoning, but it’s 75x more expensive than Haiku on output. Most day-to-day automation tasks don’t need Opus — summarisation, classification, extraction, and simple decisions all work fine on Haiku or Sonnet.
Rule of thumb: start every scenario with Haiku. Upgrade only if output quality is actually insufficient.
2. Not setting Max Tokens
Claude doesn’t know when to stop unless you tell it. Without a max_tokens cap, it will happily generate 4,000 tokens of output when you needed 200.
In Make.com, every Claude module has a Max Tokens field. Always set it. I use 512 for short responses, 1024 for medium, 2048 only when I genuinely need long output.
3. Passing the full conversation history every call
This one destroyed a client’s bill. They had a chatbot that kept appending every previous message to every new call. By message 30, they were sending 30,000 tokens of history with every response. That’s 30,000 input tokens per message × 100 messages/day × 30 days = 90 million input tokens per month. At Sonnet rates, that’s ~$270 just for input.
The fix: either truncate history (keep only last 5 messages), summarise old conversations, or use Claude’s prompt caching (cached tokens cost 10% of regular input).
4. Forgetting to deduplicate
I once built a scenario that sent the same customer email to Claude three times — once for sentiment, once for classification, once for reply. 3× the input cost for no reason. Single prompt that returns structured JSON with all three outputs = 1× the cost.
Free Tier and Credits
Is there a free tier for the Claude API? No. The Claude API is pay-as-you-go only. You need to add at least $5 credit before you can make a single call.
This is different from Claude.ai (the chat interface), which has a free tier. Same account, separate billing.
New account bonuses: Anthropic sometimes runs promotional credit for new accounts — e.g. $5 free on signup. These come and go. Check the Console when you sign up.
How to Keep Costs Predictable
Five things I do on every production workflow:
- Set Max Tokens on every single module. No exceptions.
- Start on Haiku, upgrade only when quality demands it.
- Use prompt caching for repeated system prompts or conversation context (cached tokens are 90% cheaper).
- Monitor the Usage page in the Anthropic Console weekly — it shows cost per model, per day.
- Set a hard monthly spend limit in the Console (Billing → Plans → Usage limits). Anthropic will pause the API if you hit it, which is much better than finding out next month.
Budgeting for a New Project
Rough ranges for different workflow sizes:
| Workflow volume | Monthly Claude cost (Haiku) |
|---|---|
| 100 runs/month (testing) | Less than $1 |
| 1,000 runs/month (small business) | $1–$10 |
| 10,000 runs/month (growing ops) | $10–$100 |
| 100,000 runs/month (scale) | $100–$1,000 |
These assume roughly 1,000 input + 200 output tokens per run. Your exact workflow will vary — but this gives you a planning bracket.
Next Steps
If you want to estimate exactly what a specific scenario will cost, I’d walk through it in the course. The Implementation Blueprint ($29) covers cost calculation and optimisation for real-world Claude + Make.com workflows, including a spreadsheet template for projecting monthly spend.
If you just want to try it and see, the free Quick Start walks through one scenario end-to-end, and you’ll spend maybe 10 cents of API credit running through it.
Either way — costs don’t have to be a mystery. Once you’ve built one workflow, the pricing starts to feel predictable.
Questions on your specific use case? Reply to hello@theaigeneration.co.
Last updated: 20 April 2026. I keep this page current as Anthropic’s pricing changes.
All 3 courses + AI Playbook — $49
Everything: QuickStart, Implementation Blueprint, and the AI Automation Playbook (reference PDF with prompt templates, cost calculator, and multi-API routing patterns). One payment, lifetime access.