Tokens & pricing

When you run an agent, 4RGE measures work in tokens — small chunks of text the model reads and writes. That measurement rolls up to usage for your organization. You do not need to count tokens by hand; this page explains what drives cost so daily choices feel predictable.

What this page is for

You want to…	This page helps you…
Understand "tokens"	Plain language — not a billing spreadsheet
Know what you are charged for	Agent runs, models, and context — not "per click"
See why runs vary in cost	Short fix vs long cloud job
Find admin tools	Where budgets and dashboards live
Use agents without surprise bills	Habits that match how usage actually works

Part of 4RGED Learn

Pair this with How AI models work and Context. Admins manage caps and exports in Usage & billing.

Who this page is for

You are…	Focus on
Developer / member	What affects cost → Habits
Team lead	Share habits; track spikes by project after releases
Finance / admin	How your org handles billing → Usage
New to AI tools	What is a token first

What is a token

A token is a small piece of text the model processes — part of a word, a symbol, or punctuation. Your prompt, the agent's reply, tool output, and system instructions all become tokens.

Idea	Plain meaning
Not "one message = one charge"	Cost follows how much text moved through the model
Rough size	~100 tokens ≈ a short paragraph of English (varies by language)
You do not tokenize manually	4RGE meters usage automatically on each run

Think of tokens like metered electricity: the agent run is the appliance; tokens are the units on the meter.

Input vs output

Usage usually splits into two buckets:

Type	What it is	Example
Input tokens	Everything sent to the model	Your prompt, open files, chat history, tool results
Output tokens	Everything the model generates	Replies, proposed diffs, plans

Output often costs more per token than input on many providers — long agent answers and big patches add up faster than a short question.

Back-and-forth in one run

A single agent task with many tool steps still accumulates tokens across the whole run — not as separate "message fees," but as growing input and output volume.

What counts as usage

In 4RGE, usage generally comes from agent work that calls a model:

Activity	Typically metered?
Desktop agent chat	Yes — model calls in the run
CLI agent runs	Yes
Cloud agents	Yes — often longer runs, more tokens
Approving a command	The approval itself is not a separate token line item — the model steps around it are
Reading docs offline	No model call — no token usage

Exact breakdowns (by model, project, surface) appear in Usage & billing for admins. Members may see soft limits or messages when org policy blocks a run.

What affects cost

Factor	Why it matters
Model choice	Heavier models usually cost more per token — see How AI models work
Run length	Multi-step agents read more files and produce more output
Context size	Large threads and many `@` attachments increase input tokens — Context
Task type	"Fix this test" < "Refactor five services"
Surface	Cloud runs that continue for hours can exceed a quick desktop fix

Right-size the model

Use a fast default for small edits; reserve capable models for planning and multi-file work. Admins can narrow the catalog in Model control so teams do not over-pay by accident.

How your org handles billing

4RGE usage is organization-scoped:

Layer	Who sets it	What members feel
Plan / contract	Your company + platform operators	What is included each period
Usage dashboard	Admins	Trends, exports, denials
Budget policies	Admins	Warnings or blocks when caps are hit
Model catalog	Admins	Which models appear in pickers

Included allowance is what your agreement covers; on-demand (if enabled) is spend beyond that. Members consume allowance when they run agents — they do not need the full dashboard to work, but they should know when a run was blocked for budget.

Admins: see Usage & billing for caps, filters, and CSV export. Enterprise contract detail lives in the Console — separate from day-to-day member work.

Habits that keep spend predictable

One clear task per run — less context churn, fewer wasted tokens.
Plan before large edits — avoids expensive false starts.
Pick the lighter model when the task is small.
Start fresh when a thread is huge — compressed history still consumes context.
Watch CI and cloud automations after changes — they can spike org usage without anyone noticing in chat.

Leads: after a release, ask admins to filter Usage by project or surface if spend jumped.

Where to go next

Your goal	Next step
Context and window size	Context
Model basics	How AI models work
Admin dashboards	Usage & billing
Allow models & defaults	Model control
Org vocabulary	Core concepts

Questions about tokens and pricing?

Question	Answer
Do I pay per message?	No — usage follows tokens (input + output) across the run.
Why was my run blocked?	Often budget or model policy — ask an admin; see Usage.
Does a longer chat always cost more?	Usually yes — more history and tool output increases input tokens.
Who sees org spend?	Typically admins; members may see limits in-app.
Will a smarter model always use more?	Generally yes per token; a short run on a fast model can still beat a long run on a heavy one.
Where is my commercial plan changed?	Not in Usage — that is contract / operator territory (Console).

Context

What fills the window — and why long threads cost more.

How AI models work

Models, pickers, and org catalogs.

Usage & billing

Dashboards, budgets, and exports for admins.

Model control

Defaults and allowed models per surface.

Core concepts

Org, project, usage — how pieces connect.