# How usage is counted

> Tokens, the rolling 5-hour window, the monthly limit, model multipliers, what uses quota and what does not, warnings, and what happens when you run out.

RoGenie measures usage in **tokens**, the small pieces of text models read and write. Your plan sets two limits, and both apply at once.

## The two limits

| Limit | How it works | When it frees up |
|---|---|---|
| **5-hour window** | Everything you used in the **last 5 hours**. It is a rolling window, not a fixed daily reset. | Gradually, as older usage ages out. The time shown is when your oldest usage in the window ages out. |
| **Monthly limit** | Everything you used in your current **monthly cycle**. | At the start of your next cycle: the 1st of the month on Free, and on paid plans the day of the month your subscription started (midnight UTC). |

For example, on Builder you can use 3M tokens in any 5-hour stretch and 30M in the month. See every plan in [Plans and limits](https://rogenie.ai/docs/plans).

## Tokens are counted at the model's multiplier

What counts against your quota is **billable tokens**: the tokens a model read and wrote, weighted by that model's price relative to Claude Opus 5, which is 1×.

- Input (what the model reads: your message, the conversation so far, tool results) and output (what it writes) are weighted **separately**, by that model's own input and output rates. At the 1× reference rate, an output token costs five times an input token.
- When the model supports prompt caching, the part of the conversation it has already read is billed at a lower cached rate.
- The multiplier label in the [model picker](https://rogenie.ai/docs/models) is an estimate for a typical run; hover it to see the model's exact input and output rates. A 0.1× model makes the same quota last about ten times as long as Opus 5.
- A model with no published price is billed at the default 1×.

To see what one reply cost, click **Info** on it: `quota_billable_tokens` is what your quota counted for that run.

## What uses quota

Every model call made for you, at the multiplier of the model it ran on:

- the agent's steps in a run, including playtests and checks that call the model,
- [sub-agents](https://rogenie.ai/docs/sub-agents) and [code reviews](https://rogenie.ai/docs/code-review),
- [summarizing](https://rogenie.ai/docs/chat-memory) a chat.

Long chats cost more per step, because every step re-sends the conversation. Summarizing and starting new chats for new topics keeps this down.

## What does not use quota

- Models running on [your own API keys](https://rogenie.ai/docs/own-api-keys) or [your own Claude accounts](https://rogenie.ai/docs/claude-accounts). They also keep working when your RoGenie quota runs out.
- Custom mascot generations, which have their own per-plan limit per 5-hour window. One small exception: when a vision model reads a reference image you attached, that short call counts as usual.
- Browsing the app, the IDE, chat search and settings.

## Where to see your usage

- **The library**, at the bottom: your plan and what is left for the last 5 hours and the month.
- **Settings › Account & Usage**: cards for *Last 5 hours*, *This month* and *Mascots (5h)*, each with used, left and when it frees up.
- **Info** on any reply, for that run.

## Warnings before you run out

When you reach **75%** of either limit, RoGenie tells you and offers to switch the chat to **Auto (intelligent & cheap)**, which runs a capable, low-cost model at a fraction of the quota.

Before each model call, RoGenie also checks that the step fits in what you have left, so a single large step cannot overshoot your limit.

## When you run out
When what is left is not enough for another step, RoGenie stops the run and says which limit you reached, for example:

- *5-hour token limit reached (2,995,000 / 3,000,000 tokens): what is left is not enough for another step. This is a rolling 5-hour window, so it frees up as older usage ages out.*
- *Monthly token limit reached (...): what is left is not enough for another step. It resets at the start of your next monthly cycle.*

Built-in models then stop until the limit frees up. The reply shows when, plus your options:

1. **Wait.** The 5-hour window frees up gradually; the reply shows the time.
2. **Use your own keys or Claude accounts.** Models on [your own API keys](https://rogenie.ai/docs/own-api-keys) (any plan) and your connected [Claude accounts](https://rogenie.ai/docs/claude-accounts) (Builder and up) keep working, and their usage never counts toward your RoGenie quota.
3. **Invite friends.** [Referrals](https://rogenie.ai/docs/referrals) add bonus quota.
4. **Upgrade.** The card offers plans with higher limits. An upgrade applies right away.

Everything RoGenie already did in Studio stays. When your quota frees up, ask it to continue: it picks up from the place's current state instead of starting over.

## Changing plans

Changing plan does not erase usage you already have. A higher plan's limits apply as soon as the upgrade goes through; a downgrade applies at the end of your billing period. See [Upgrade, downgrade and cancel](https://rogenie.ai/docs/billing).

## Make your quota go further

- Use **Auto (intelligent & cheap)** or another low-multiplier model for everyday work.
- [Summarize](https://rogenie.ai/docs/chat-memory) long chats, and start a new chat for a new topic.
- Keep **Efficient** pacing and **Smart context compression** on in [Agent settings](https://rogenie.ai/docs/agent-settings).
- Use a lower [effort](https://rogenie.ai/docs/effort-and-fast-mode) for simple tasks, and avoid Fast mode unless you need it.
- Pick cheap models for simple [sub-agent roles](https://rogenie.ai/docs/sub-agents#sub-agent-models).
- For small, low-risk changes, turn off [playtest verification](https://rogenie.ai/docs/agent-settings#playtest-verification).