Prices checked 23 October 2026. AI model prices change often, usually downwards for a given capability. Use the estimator with current prices from your provider.
When you add an AI step to an automation (classify an email, draft a reply, extract invoice fields), you pay the model provider for usage. The pricing looks strange at first (dollars per million tokens), but it’s easy to estimate.
Tokens in plain terms
- A token is a chunk of text, often part of a word. As a rough guide for English, 1,000 tokens is around 750 words. Arabic and other languages can use more tokens for the same meaning.
- Input tokens: everything you send: instructions, your FAQ or documents, the customer’s message, the conversation so far.
- Output tokens: what the model writes back.
- Output tokens cost several times more than input tokens.
What prices looked like (October 2026)
Published API prices from OpenAI and Anthropic fell into similar tiers:
| Tier | Input (per million tokens) | Output (per million tokens) | Good for |
|---|---|---|---|
| Efficient / small models | around $0.10 | around $0.50 | Classification, extraction, short replies at volume |
| Mid-tier models | around $2 | around $10 | Drafting, summarising, most business tasks |
| Most capable models | around $4–$10 | around $20–$50 | Complex reasoning, long documents, agents |
Providers also offer discounts: cached input (repeated instructions or documents) costs much less, and batch processing (results within hours rather than seconds) is often around half price.
A worked example
An AI step drafts replies to 1,000 customer messages a month. Each call sends about 2,000 input tokens (instructions + FAQ + message) and gets back 300 output tokens.
- Input: 1,000 × 2,000 = 2 million tokens
- Output: 1,000 × 300 = 0.3 million tokens
| Tier | Monthly cost |
|---|---|
| Efficient (0.10 / 0.50) | 2 × 0.10 + 0.3 × 0.50 = $0.35 |
| Mid-tier (2 / 10) | 2 × 2 + 0.3 × 10 = $7 |
| Most capable (10 / 50) | 2 × 10 + 0.3 × 50 = $35 |
For most small-business uses, AI usage is a small line compared with the time saved. Costs climb with very long context (whole documents in every call), agents that make many calls per task, and high volumes.
What drives the bill up
- Sending too much context: the full product catalogue in every request instead of the relevant part
- Long conversation history resent on every turn
- Agents that loop through many tool calls per task
- Using the top model for everything, including simple classification
- Retries and errors that repeat calls
Keeping costs under control
- Match the model to the task: efficient models for classify/extract; stronger ones only where quality needs it.
- Send only what’s needed: retrieve the relevant FAQ entries, not all of them.
- Use caching for repeated instructions and documents where your provider supports it.
- Batch non-urgent jobs.
- Set spending limits and alerts in the provider’s console.
- Log tokens per run in your automation so you can spot outliers.
Remember the platform costs too: automation runs (n8n executions, Make credits, Zapier tasks). See what automations really cost.
Frequently asked questions
Is a ChatGPT or Claude subscription the same as API usage?
No. Subscriptions are per-person chat apps with usage limits. API usage is billed per token and is what automations use.
Which provider is cheapest?
Prices for comparable tiers are often similar and change frequently. Choose on quality for your task, data terms (see using AI with business data safely) and integration, then compare price.
How do we estimate tokens for our task?
Run 20 real examples, read the token counts from the API response or your automation logs, and use the average in the estimator.
If you’d like help choosing models and estimating costs for an AI workflow, that’s part of our tech stack service.
Free download · Excel calculator
AI API cost estimator
Enter tasks per month, tokens in and out per task and the prices from your provider's current page. The sheet estimates monthly cost per task and compares model tiers.
Download itNo sign-up. Opens in Excel, Google Sheets or Numbers.
Rather have it done?
This is the kind of work our tech stack service covers. Tell us where you are today and we’ll reply with a plan and a fixed price.
Let’s talk