---
title: "AI chat for employees: seat licence or pay per token"
description: "How to compute, per person, whether an AI chat seat licence or paying per token is cheaper, with the break-even formula and a worked example."
url: https://www.cureman.de/en/articles/ai-chat-seat-licence-or-pay-per-token
language: en
updated: 2026-11-16
alternate_de: https://www.cureman.de/de/artikel/ki-chat-sitzlizenz-oder-token.md
---

# AI chat for employees: seat licence or pay per token

> A seat licence costs the same whether its holder sends five messages a month or five hundred. Paying per token costs what each person uses. Which is cheaper is a question per person with a computable answer. The break-even is the monthly seat price divided by the average cost of one of that person's requests, where a request costs its input tokens times the input price plus its output tokens times the output price. People below the break-even are cheaper on tokens, people above it on a seat, and most organisations have both.

A seat licence for an AI chat product costs the same whether its holder sends five messages a month or five hundred. Paying per token costs what each person actually uses. Which is cheaper is a question per person, and it has a computable answer: the break-even is the monthly seat price divided by what one of that person's requests costs. Below it, tokens are cheaper. Above it, the seat is.

## Two ways to pay for the same answers

| Question | Seat licence | Pay per token |
| --- | --- | --- |
| What is billed | A fixed price per person per month | Input and output tokens of every request |
| Someone who never signs in | Full price | Nothing |
| Someone who works in it all day | Same price, within the vendor's usage limits | Rises with every request |
| What comes with it | The vendor's application and its features | A model behind an API; the chat front end is yours to provide |
| What you run | User management in the vendor's console | Front end, sign-in, keys, budgets |
| When cost can be reduced | Usually at renewal | From the next request |

## What one request costs

```
cost of a request = (input tokens × input price + output tokens × output price) ÷ 1,000,000
```

Prices are quoted per million tokens, and output costs more than input. The input of a chat request is the whole conversation so far, so a message late in a long chat costs more than the first one. Use the average over real requests, not a short test question. The parts of the formula are explained in [LLM token costs](https://www.cureman.de/en/articles/llm-token-costs-explained).

## The break-even per person

```
break-even requests per month = seat price per month ÷ average cost per request
```

A worked example with invented figures. The seat costs 30.00 a month. The model costs 4.00 per million input tokens and 20.00 per million output tokens. An average request has 3,000 input tokens and 500 output tokens.

- Cost per request: 3,000 × 4.00 ÷ 1,000,000 + 500 × 20.00 ÷ 1,000,000 = 0.012 + 0.010 = 0.022
- Break-even: 30.00 ÷ 0.022 = about 1,360 requests a month

| Person | Requests a month | Token cost | Seat | Cheaper |
| --- | --- | --- | --- | --- |
| Occasional user | 150 | 150 × 0.022 = 3.30 | 30.00 | Tokens |
| Regular user | 600 | 600 × 0.022 = 13.20 | 30.00 | Tokens |
| Heavy user | 2,500 | 2,500 × 0.022 = 55.00 | 30.00 | Seat |

## From one person to the organisation

The same invented figures for 100 people, of whom 60 are occasional, 30 regular and 10 heavy users:

| Option | Calculation | Per month |
| --- | --- | --- |
| A seat for everyone | 100 × 30.00 | 3,000.00 |
| Tokens for everyone | 60 × 3.30 + 30 × 13.20 + 10 × 55.00 | 1,144.00 |
| Seats for the 10 heavy users, tokens for the rest | 10 × 30.00 + 198.00 + 396.00 | 894.00 |

The decision is made per person, not once for the company. How your people spread across the three rows is the thing to measure. The example says nothing about it.

## Where the four inputs come from

1. **Seat price.** What you pay per user per month according to the contract or the invoice, not the list price. Note the billing period and any minimum number of seats.
2. **Requests per person.** For existing seats, the usage export of the vendor's admin console, where it reports messages per user. For per-token access, the log of whatever sits between the front end and the provider.
3. **Tokens per request.** A seat product does not report them. Until a group has worked through per-token access for a month, this is an assumption. Write it down as one.
4. **Model prices.** The provider's price list for the model people would actually use, on the day you compute.

## What the calculation leaves out

- **Features of the seat product.** File analysis, image generation, voice, connectors to other tools, mobile apps. A chat front end on an API may offer fewer. List what people use before comparing prices.
- **Usage rises when nothing rations it.** A person who was careful with a capped seat may send more once every request simply goes through. Compute again after two months of real use.
- **The model is now a cost decision.** Per token, the most capable model is also the most expensive on every request. A list of allowed models and a monthly budget per person keep that in hand, as described in [AI budgets and limits](https://www.cureman.de/en/articles/ai-budgets-and-limits).
- **A margin.** Do not move someone at the exact break-even. Move the people who are clearly below it and leave the ones near it where they are.
- **The contract.** Seats can usually be reduced only at renewal, so the date that matters is the notice date. See [preparing a SaaS renewal](https://www.cureman.de/en/articles/prepare-saas-renewal).
- **Attribution.** The comparison needs usage per person. How to record it, and what to settle with employee representatives first, is covered in [allocating AI costs to teams](https://www.cureman.de/en/articles/allocate-ai-costs-to-teams).
- **Terms of use.** What a provider may do with prompts can differ between its chat product and its API. Read both.

## How Cureman does it

The AI workspace of Cureman is a chat front end for employees, connected to its gateway with a key of its own. Open WebUI is the default, and any front end that speaks the OpenAI chat format works the same way. The front end passes on who is signed in, so every request is attributed to a person. People choose from the models the organisation allows, and a monthly budget per person can be set. The provider bills the tokens to the organisation directly, on the organisation's own API key.

Each month the comparison is shown as arithmetic: the people who used the workspace, counted as everyone with at least an agreed number of requests in the month; what a seat for each would have cost; the provider cost of their tokens; and the difference as the saving. The seat price is a figure agreed in the terms and labelled as that, not a measurement of what was paid. A month in which the tokens cost more than the seats would have shows no saving, and nothing is charged for the module. Requests that name no user count towards the cost and for no seat.

The same comparison runs in both directions elsewhere in the product. For chat seats an organisation already pays, the licence module lists holders whose recorded use would clearly cost less per token, with the tokens assumed per message stated as an assumption. For people whose token cost on certain tasks is clearly above a seat, prompt insights recommends the seat. The module is described on the page about the [AI workspace](https://www.cureman.de/en/modules/ai-workspace).

## Questions and answers

### Is paying per token cheaper than a seat licence for AI chat?

It depends on the person. Someone who sends a few messages a week costs far less per token than a seat, and someone who works in the chat all day can cost more. Compute the break-even per person from the seat price and the average cost of a request, and expect to end up with a mix.

### How do I calculate the break-even between a seat and tokens?

Compute what one average request costs, as input tokens times the input price plus output tokens times the output price, divided by one million. Divide the monthly seat price by that cost. The result is the number of requests a month at which both ways of paying cost the same.

### How many tokens does one chat message use?

There is no fixed number. Every message sends the whole conversation so far as input, so the tenth message of a chat uses far more input tokens than the first. Until you have measured it for your own people, any tokens-per-message figure is an assumption and should be labelled as one.

### What is needed to offer employees AI chat billed per token?

An API account with a model provider, a chat front end that speaks to it, sign-in with the company's identity provider, and something in between that records who used what, restricts the models on offer and enforces a monthly budget per person. Without attribution per person the comparison with seats cannot be made.

### How does Cureman compute the saving of its workspace?

For each month it counts the people who used the workspace, multiplies them by the seat price agreed as the comparison, and subtracts the provider cost of the tokens those people used. The seat price is labelled as an agreed figure, not a measurement. A month in which the tokens cost more than the seats would have shows no saving and is not charged.

## The module that does this

[AI workspace: an employee chat paid by the tokens used](https://www.cureman.de/en/modules/ai-workspace): The AI workspace is a chat front end for employees on the AI gateway. You pay the provider for the tokens used instead of a seat licence for everyone.

## Read next

- [LLM token costs explained: input, output and cached tokens](https://www.cureman.de/en/articles/llm-token-costs-explained): What input, output and cached tokens are, why output tokens cost more than input tokens, and how to compute the cost of a single LLM request step by step.
- [AI budgets and limits: per key, per user, per model](https://www.cureman.de/en/articles/ai-budgets-and-limits): How monthly budgets per API key and per user and model allow-lists work for LLM spend, what happens at the limit, and how to set the amounts.
- [How to allocate AI costs to teams, projects and people](https://www.cureman.de/en/articles/allocate-ai-costs-to-teams): A method to attribute LLM spend to teams, features and people with keys, tags and user identifiers, price every request and reconcile with the provider.
