Quick answer
Auto mode lets Cursor choose a model, usually at a lower or included rate. Selecting a specific premium model bills that model's rate, which can be several times higher. That is why two identical prompts can consume very different amounts of your allowance.
The difference
| Mode | Model choice | Cost behaviour |
|---|---|---|
| Tab completion | Cursor's fast model | Included or minimal |
| Auto | Cursor picks | Lower, often included |
| Premium selection | You pick | Priced per model |
| Agent on premium | You pick, many calls | Highest |
Read the split
The dashboard shows premium consumption separately from Auto. If the premium line is most of your spend, the decision is model choice. If Auto dominates, the decision is volume or agent loop length.
- Premium share high: reserve strong models for hard tasks.
- Auto share high: shorten sessions and reduce retries.
- Both high: you are running long agents with large context.
- Neither high but you are blocked: check the billing period.
Use the split to change one thing
- Start a week with Auto as the default.
- Switch to a premium model only after Auto fails once.
- Note which tasks actually needed the premium model.
- Keep the short list and drop the rest.
UsageAtlas keeps Cursor's reported model split next to your other tools' token splits, so the comparison is consistent.
Use the split to change one habit
The model split is diagnostic, not decorative. If premium models dominate, the fix is model choice. If Auto dominates, the fix is session length and retry behaviour. Trying both at once makes the result unreadable.
- Premium-heavy: reserve strong models for hard tasks.
- Auto-heavy: shorten agent runs and cap retries.
- Balanced: compare per task rather than per request.
- Cost rising with flat volume: check context size.
See Cursor usage for where the numbers come from and when limits reset for planning.
Questions, answered
Is Auto mode free in Cursor?
It is treated far more generously than premium models and is often included, but the exact rules depend on your plan. Check the plan page.
Why is one premium request so expensive?
Premium models are priced per model and the request includes the whole context. An agent turn can make several calls.
Can I see which model Auto chose?
Cursor shows the model in the request details where available. The dashboard aggregates rather than listing every request.