Factory Droid token usage is not billed token for token: Droid turns the tokens a model processes into Factory Standard Credits, scaled by a multiplier that differs from model to model, and those credits fill your rolling limits. For a record of what you ran, use Droid's own /stats or /cost; to see how close the limits are to full, add the free UsageAtlas app to your desktop.
Think of it like a game with two meters. One counts how much you played. The other shows how much energy is left before you have to wait. Droid has both, and they sit on different screens.
Tokens, credits and multipliers
Every request to a model sends input tokens and gets output tokens back. Factory does not charge your plan in raw tokens. Its docs say Droid sessions consume Factory Standard Credits based on model usage, along with compute when you use Droid Computers.
The models page lists a multiplier for every model. A high multiplier means the same amount of work costs more credits; a low one means less. The Droid Core models, Factory's open-weight pool, sit in their own section of that list, and many carry low multipliers.
| Term | What it counts | Where you see it |
|---|---|---|
| Usage statistics | What your sessions used over a period | /stats, /cost |
| Context | How full the current session's window is | /context |
| Standard Credits | Tokens turned into plan usage via the model multiplier | Settings > Usage, /limits |
| Rate Limits | 5-hour, 7-day and 30-day rolling caps on credits | /limits, UsageAtlas |
Reading your token usage inside Droid
Factory's CLI reference describes /stats [period] as usage statistics that accept relative periods and date ranges, and /cost as usage statistics too. Use /stats when you want last week or a custom range. /context is different: it shows how much of the current context window a session is using, with a progress bar. A big context means every new turn resends a lot of tokens.
Why the same task can cost very different amounts
Two things move your credit spend more than anything else. The first is the model. A frontier model with a high multiplier drains Standard Usage faster than a Core model doing the same edit. The second is context. A session that has grown long sends more input tokens per turn, and that adds up across a 5-hour window.
So the practical habits are simple: run routine work on a lighter model, start fresh sessions for new tasks, and keep heavy models for the hard parts.
Watching the limits, not just the tokens
Token counts tell you what happened. The limits tell you whether your next request will go through. UsageAtlas is a free, open-source desktop app for Windows, macOS and Linux that connects to Factory with an API key and shows the Standard and Core windows separately, along with extra balance and reported resets. You can set an alert on a supported window.
One honest note: for Factory, UsageAtlas reads limits and balances, not per-request tokens. If you mainly want a history of past sessions, Droid's /stats is the right tool. UsageAtlas does keep token history for tools that write it locally, such as Claude Code and Codex, so it is the better pick when Droid is one of several agents you use.
Questions
Why did my credits drop fast on a short task?
Usually a high-multiplier model or a long context. Try the same task on a Droid Core model and compare.
Can I see token usage in the Factory App?
The docs point to Settings > Usage for plan usage. Usage statistics by period are in the CLI's /stats.