What AI Really Costs a Small Firm: Tokens, Seats and the Bills Nobody Saw Coming
Most small firms pay for AI twice: once in per-person seats, and again in usage nobody is watching. Tokens are cheap one at a time, but long documents, retries and agents that loop add up quietly. Here's what each piece costs, with current prices and a calculator for your own bill.
What an AI token costs, in plain dollars
A token is the unit AI models read and write in. It's a chunk of a word: in English, one token is roughly three-quarters of a word, so 1,000 words is about 1,300 tokens. API providers charge per million tokens, with separate prices for what you send in (input) and what the model writes back (output). Output usually costs four to five times more than input.
Some worked examples, using the API prices in the table below. Your numbers will differ with the length of your documents and the model you pick.
- Summarising 100 emails. At about 400 words each, that's roughly 53,000 tokens in and 13,000 tokens out for short summaries. On a mid-range model at $2 in and $10 out per million tokens, that's about $0.24. On a small, cheap model at $0.10 and $0.50, it's about one cent.
- Reviewing a 50-page contract. About 25,000 words is around 33,000 tokens in, plus a 750-word summary out. On the same mid-range model that's about $0.08.
- An agent working through a task in 20 steps. If it re-reads a 30,000-token context at every step, that's 600,000 input tokens for one task, about $1.20 on the mid-range model. Run that 200 times a month and it's around $240. This is where usage bills surprise people.
Subscriptions vs API: seats are the bill you can see
Most small firms meet AI as a monthly seat: a fixed price per person for a chat app. The API is the pay-per-token route that software and automations use. Seats are predictable but get paid whether anyone uses them. API usage is cheap for light work and unpredictable for heavy work.
Seat prices (US list prices per user per month, before tax; prices checked Oct 12, 2026):
| Plan | Billed monthly | Billed annually | Notes |
|---|---|---|---|
| ChatGPT Plus | $20 | Not offered | For individuals. OpenAI Help Center |
| ChatGPT Business (standard seat) | $25 | $20 | From 2 users. Premium seat $125 / $100. OpenAI business pricing |
| Claude Pro | $20 | $17 | For individuals. Claude pricing |
| Claude Team (standard seat) | $25 | $20 | Minimum 2 members. Premium seat $125 / $100. Claude Help Center |
| Microsoft 365 Copilot Business (add-on) | $25.20 | $21 ($18 promo) | Needs a qualifying Microsoft 365 business plan on top. The $18 first-year promo runs to Dec 31, 2026. Microsoft |
| Google Workspace Business Standard (Gemini included) | $14 | n/a | Gemini in Gmail, Docs, Sheets, Meet and more. Business Starter ($7) has Gemini in Gmail only. New-customer promos may apply. Google Workspace pricing |
API prices (US dollars per million tokens, standard tier; prices checked Oct 12, 2026):
| Model | Input | Output | Source |
|---|---|---|---|
| OpenAI GPT-6 Luna (small) | $0.10 | $0.50 | OpenAI API pricing |
| OpenAI GPT-6.1 Sol (mid-range) | $2.00 | $10.00 | |
| OpenAI GPT-6 Astra (top) | $10.00 | $50.00 | |
| Anthropic Claude Haiku 5.5 (small) | $0.50 | $2.50 | Claude pricing |
| Anthropic Claude Sonnet 5.5 (mid-range) | $2.00 | $10.00 | |
| Anthropic Claude Opus 5.5 (top) | $4.00 | $20.00 |
Prices change often and vary by region, tax and contract. Long prompts, batch processing, caching and US-only processing can be priced differently. Check the vendor's page before you budget.
AI cost calculator: your monthly bill
Enter how many seats you pay for and, if you use the API, roughly how much work goes through it. The prices are filled in from the tables above. Change them if your contract is different.
The costs nobody budgets for
- Unused seats. Seats bought in a burst of enthusiasm and opened twice since. Every one is billed every month.
- Duplicate tools. One person on ChatGPT, another on Claude, a Copilot add-on nobody opens, and AI features inside the accounting and project software too. Several tools often do the same job.
- Agent loops and retries. An automation that fails and retries, or an agent that takes 30 steps instead of 5, multiplies the token bill without anyone noticing until the invoice arrives.
- Long context. Sending a whole document or email thread every time, when a short excerpt would do, is the commonest way a cheap task becomes an expensive one.
- Staff time reviewing output. Usually the largest cost and never on the invoice. If someone spends 20 minutes checking every AI draft, that's what the work really costs.
How to audit your AI spend in an afternoon
- List every AI charge from the last three months: card statements, app stores, software invoices and API dashboards.
- Match each seat to a person and check when they last used it. Cancel or downgrade the ones nobody uses.
- Look for overlap. Pick one main chat tool for the firm unless there's a clear reason for two.
- Set a monthly spend limit on every API account and turn on usage alerts.
- For each automation, check how many tokens one run uses and how many runs fail or retry.
- Write down who reviews the output and how long it takes. That's part of the cost.
A note on Microsoft 365 Copilot
Copilot is an add-on priced per person on top of a Microsoft 365 business plan, so it only pays off for the people who live in Word, Excel, Outlook and Teams all day. Buying it for everyone at once is a common way to end up with unused seats. A fuller breakdown of Copilot costs for small firms is coming soon.
Frequently asked questions
How much does an AI token cost?
It depends on the model. On the API price lists we checked on Oct 12, 2026, input ranged from $0.10 to $10 per million tokens and output from $0.50 to $50 per million. A million tokens is roughly 750,000 words, so a single token costs a tiny fraction of a cent.
How much does AI cost per month for a small business?
For most small firms the main cost is seats, at roughly $14 to $25 per person per month on the business plans we checked, plus any API usage from automations. A 10-person firm on one business chat plan pays around $200 to $250 a month before usage and staff time.
Is a subscription or the API cheaper?
For people chatting with AI every day, a seat is usually simpler and more predictable. For automations that handle a set task many times, the API is often cheaper per task, but it needs spend limits because a looping or retrying job can run up the bill.
How do I estimate my token costs?
Count tasks per month, estimate the words going in and out of each task, divide by 0.75 to get tokens, then multiply by the model's price per million tokens. The calculator on this page does the arithmetic.
How can a small firm cut its AI costs?
Cancel unused seats, standardise on one main tool, use smaller models for simple tasks, send only the text the model needs, and set API spend limits. Then check whether the time saved is worth what you're paying.
Find where your AI spend leaks
Tell us which AI tools your firm pays for and what you use them for. You'll get a straight answer on what's worth keeping, what overlaps, and where the money goes.