Models and cost

Credits and what you're spending

See the cost of a turn as it happens, the session broken down by model, and your balance.

Code
/usage      this session by model, plus your balance
/upgrade    the billing portal — no login needed

The status line shows the cost of the current turn while it runs, so an expensive turn is visible while it's happening rather than at the end of the month.

How it's priced

Managed calls draw ContextStream credits at roughly 1.10× the provider's own rate for whichever model ran. Failed calls aren't billed.

Calls through your own key cost zero credits and bill your provider directly.

What costs the most

In rough order:

  1. Parallel orchestration — several agents, several model calls at once.
  2. Deep research — many pages read, all of it model input.
  3. Long sessions without compaction — the whole history is re-sent each turn. /compact folds it down.
  4. Image generation.

Ordinary editing turns are small by comparison.

Keeping it down

  • Pin a cheaper model for mechanical work: /model fast.
  • /compact when a session has been running for hours.
  • Use /plan before big work — a good plan avoids the expensive kind of retry.
  • Bring your own key if you already have provider credit.

If it doesn't work

A turn cost more than expected. /usage breaks it down by model. A turn that fanned out to agents is several calls, not one.

Balance is empty mid-session. /upgrade opens the portal without a login step. Configured provider keys keep working regardless.

The status line shows no cost. You're on your own key for that model — the spend is on your provider's dashboard.

Next