AI costs a business in two ways: flat-fee subscriptions for each person who uses it, or pay-per-token usage through APIs, when AI works inside software or an automation. Business subscriptions to the main assistants usually cost around $20–30 per user a month. APIs are priced per million tokens: OpenAI’s GPT-5.5 costs $5 per million input tokens and $30 per million output tokens, while the smallest models in the same family cost up to 25 times less. The most effective way to keep spending low is to use the smallest model that does the job well.
I’m Toure Tia Kevin, a developer in Bassano del Grappa, Italy. I write the software and scripts for my projects with Claude and build AI automations for businesses: API costs are something I check every month.
Subscription or API: what changes
| Per-user subscription | Pay-as-you-go API | |
|---|---|---|
| How you pay | fixed monthly fee per person | by the tokens used |
| For what | chat, writing, document analysis | automations, chatbots, software |
| Predictable cost | yes | depends on volume |
| Examples | ChatGPT Business, Claude Team, Gemini in Google Workspace | OpenAI, Anthropic, Google, Mistral |
If you use AI through the chat, you pay the subscription and API price rises don’t affect you. If you have a chatbot on your site or an automation, you pay per use, and then the choice of model matters a lot.
OpenAI API prices
Prices per million tokens from OpenAI’s official price list, November 2026:
| Model | Input | Output |
|---|---|---|
| GPT-5.5 | $5.00 | $30.00 |
| GPT-5.4 | $2.50 | $15.00 |
| GPT-5.4 mini | $0.75 | $4.50 |
| GPT-5.4 nano | $0.20 | $1.25 |
GPT-5.5 costs twice as much as GPT-5.4 per token. Text that’s already been sent and repeats (cached input) costs a tenth, and non-urgent batch processing costs half. Prices change often: always check the provider’s price list.
What’s a token? A piece of a word: in English, a thousand tokens are roughly 750 words. A 150-word reply to a customer, with the instructions and context that go with it, costs about two cents with GPT-5.5: not much, but ten thousand replies a month add up to around two hundred dollars.
How to spend less without losing quality
- The right model for each task: small models are enough for sorting emails, summarising or answering common questions. The most powerful model is for complex reasoning.
- Reuse context: instructions and documents repeated on every request cost far less when cached.
- Batch processing: reports and analyses that aren’t urgent cost half as much.
- Spending caps: every provider lets you set a monthly limit. Do it on day one.
- Measure: an automation that records what it spends per operation gets optimised quickly.
What an AI project costs
Beyond subscriptions and APIs, there’s the work of building the system. In my AI and automation service, an initial consultation or a first automation starts at €500. An AI chatbot for your website is quoted per project, depending on content and integrations. I write custom scripts with Claude rather than use automation platforms whose fees grow with use.
What about European models?
If your data has to stay in Europe, Mistral AI offers competitively priced models with servers in the EU. I cover it in the guide on Mistral AI for business.
Frequently asked questions
Does GPT-5.5 cost more than GPT-5.4?
Yes: per token the price is double, $5 against $2.50 for input and $30 against $15 for output. It’s worth using only where its reasoning ability is really needed.
If I use ChatGPT on a subscription, does my cost change?
No. The subscription has a fixed fee. Per-token prices apply to people using the APIs in software and automations.
How much does a small business spend on AI each month?
It depends on use. A few subscriptions for the people who write and analyse documents, plus the APIs for an automation or chatbot, which with small models often cost from a few euros to a few tens of euros a month.
For a cost estimate for your case, get in touch.
