How tokens work
We charge by tokens because that’s what the upstream LLM providers charge us. A “request” can use 1k or 100k tokens — billing per request either burns our margin or forces us to limit you artificially.
What is a token?
Roughly four characters of English text, or one short Russian word.
The phrase def parse_token(t): is about 5 tokens. A 200-line file is
typically 1,000–2,000 tokens.
What counts on your quota
| Direction | Counted? |
|---|---|
| Your prompt (input) | Yes |
| Files / snippets sent for context | Yes |
| Model output (the reply, including tool calls) | Yes |
| Tool results returned to the model | Yes — these are part of the next prompt |
| The IDE talking to the backend | No |
In practice, an average agent turn (small refactor) uses 3,000–8,000 tokens. A long debugging session can use 50,000+.
Plans at a glance
| Plan | Monthly tokens | Crypto price |
|---|---|---|
| Trial | 250,000 (one-time) | $0 |
| Plus | 1,500,000 / month | ~$15 |
| Pro ⭐ | 5,000,000 / month | ~$35 |
| Ultra | 20,000,000 / month + Max-Quality always on + priority | ~$99 |
| Top-up | +500,000 (valid 90 days) | ~$5 |
Yearly plans get −12%. See sirius-ide.com/pricing for live numbers.
When does Ultra make sense?
Ultra is the “almost-unlimited” tier (think: Cursor Ultra, but designed for CIS prices). Pick Ultra if any of these describe you:
- You run the agent on big projects daily and constantly hit the Pro cap toward month-end.
- You want Max Quality mode enabled by default (premium reasoning models: Claude Opus 4.7, GPT-5.2-codex, o3) instead of toggling it per-task.
- You want top-priority queue placement so the agent doesn’t wait behind other users at peak hours.
Rollover policy
Monthly tokens do not roll over. Top-up packs are valid for 90 days. If you consistently leave tokens on the table, switch to a smaller plan + top-ups.