> For the complete documentation index, see [llms.txt](https://docs.rubica.au/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.rubica.au/rubica-chat/understanding-usage-and-cost.md).

# Understanding Usage and Cost

Rubica Chat is not a flat-fee unlimited tool. It works on a pay-per-use model — similar to how you might pay for data on a phone plan, except here you are paying for the amount of text the AI reads and writes. This page explains how that works and what affects your usage.

### What is a token?

A token is just a small chunk of text — roughly three to four characters, or about three-quarters of a word. Every time you use Rubica Chat, the AI is counting tokens in both directions:

* **Input tokens** — everything the AI reads: your message, any client data it retrieves, instructions from the system, and descriptions of the tools it has available.
* **Output tokens** — everything the AI writes back to you: its response, any document it generates, any data it looks up and returns.

Output tokens cost more than input tokens.

### How much does it cost?

Rubica Chat primarily uses two AI models. The costs below are the underlying rates charged by the AI providers:

| Model             | Input cost                   | Output cost                   | When it is used                                                          |
| ----------------- | ---------------------------- | ----------------------------- | ------------------------------------------------------------------------ |
| **Claude Sonnet** | USD $3.00 per million tokens | USD $15.00 per million tokens | Document generation, complex tasks, anything requiring careful reasoning |
| **Gemini Flash**  | USD $1.50 per million tokens | USD $9.00 per million tokens  | Quicker tasks, lookups, summarisation                                    |

**To put that in plain terms:** one million tokens is roughly 750,000 words — about ten novels. A typical conversation uses a tiny fraction of that. A short Q\&A might cost less than a cent. However, generating a full SOA — which involves retrieving client data, working through a long template, and producing thousands of words of output — costs meaningfully more. Think dollars per document, not fractions of a cent.

> Usage costs flow through to your firm's Rubica subscription. Your administrator can provide information on how usage is tracked and billed for your firm.

### What makes usage go up?

The biggest factors that drive token consumption are:

#### The tools you have switched on

Every tool (connection) you have active adds its own description to the AI's context — before you have even typed anything. The more tools that are switched on, the more the AI is reading at the start of every single message. Having five tools on when you only need one is like sending the AI a five-page brief when you only needed to ask one question.

**What to do:** only switch on the tools you actually need for the current task.

#### Retrieving client data

When the AI pulls a client's fact find, risk profile, or RetireMap modelling, all of that data gets loaded into the conversation. A detailed fact find can be quite large. This is unavoidable for tasks that need it, but it is worth knowing it adds to the total.

#### How long the conversation runs

The AI re-reads the entire conversation every time you send a new message. So a very long conversation — with lots of back and forth — means the AI is reading more and more with each message. This is another reason to start a fresh chat for each new client or task.

#### How much the AI writes

A short answer costs less than a long one. Generating a full SOA produces thousands of words of output, which is why document generation tasks cost more than a quick lookup. That is expected — it is just useful to understand.

#### Which model is being used

Claude Sonnet is more capable but costs more. Gemini Flash is faster and cheaper, and works well for simpler tasks. Using the right model for the job makes a real difference over time.

### Common misconceptions

**"It's like a subscription — I can use it as much as I want."** No. Usage is metered. Think of it more like a utility bill than a Netflix subscription.

**"It doesn't matter how many tools I leave on."** It does matter. Each active tool adds overhead to every single message in the conversation. Switching off tools you are not using is one of the easiest ways to reduce unnecessary cost.

**"Starting a new chat doesn't help."** Actually it does. The AI re-reads the full conversation history with every message. A fresh chat means it starts clean, without all that accumulated history to re-read.

**"Shorter replies from the AI save money."** Partly — but the cost of input (what the AI reads) often matters just as much as the cost of output (what it writes), especially when multiple tool calls are happening in the background.

### Simple habits that keep usage efficient

* Start a fresh chat for each client or each distinct task.
* Only enable the tools you need for what you are doing right now.
* Use Gemini Flash for quick lookups and simple questions; use Claude Sonnet when you need it to think carefully or generate a long document.
* Be clear and specific in your prompts — a vague question that the AI has to clarify back and forth costs more than a clear one that gets the answer right the first time.