Last updated: July 2026
Cursor Auto mode lets Cursor pick a cost-efficient model for each request, and that usage is included at no extra cost — it does not draw down your ~$20 monthly credit pool. Manually selecting a frontier model like Claude Opus or GPT-5 is what spends the pool, and overages after it runs out bill at API rates.
Auto is one of Cursor's most misunderstood settings. Used well, it handles the vast majority of your coding for free within your subscription and keeps your credit pool in reserve for the moments that actually need a frontier model. Here is exactly how Auto works, how it differs from manually picking a model, and when it is worth overriding.
When you open Cursor's model picker, Auto sits at the top of the list. Instead of committing every request to one specific model, Auto lets Cursor choose a cost-efficient model for each request on your behalf. For a small edit it reaches for something lightweight; for a slightly meatier task it may route to a stronger model — but always within the efficient tier rather than the frontier tier.
The key property that makes Auto special is billing: usage in Auto mode is included at no extra cost. It does not draw from the monthly credit pool that comes with your Cursor Pro subscription. In practice that means you can work all day in Auto and never touch the credits you are saving for hard problems. For the full breakdown of tiers and credits, see our Cursor pricing guide.
The difference between Auto and manually selecting a model is not just which model runs — it is where the money comes from. This is the single most important thing to understand about Cursor's cost model.
In Auto mode, requests are covered by your subscription and do not spend the credit pool. In manual mode, the moment you pin a frontier model — Claude Opus, GPT-5, or another top-tier model — every request draws from your roughly $20 monthly credit pool. Big requests on large contexts drain it quickly.
When that pool is exhausted, Cursor does not stop working. You have two paths: fall back to Auto and keep going at no extra cost, or stay on the frontier model and pay overages at API rates for every token until the month resets. There is no automatic downgrade, so a forgotten frontier default is exactly how bills spike. Our guide on why Cursor gets expensive walks through how that plays out across a month.
Because Auto is included, you rarely need to think about its token pricing at all — the point is that it does not spend your pool. But it helps to know roughly what the efficient models under Auto are priced at, because it shows just how large the gap is when you jump to a frontier model.
| Token type | Approx. Auto rate |
|---|---|
| Cache read | ~$0.25 / M tokens |
| Input | ~$1.25 / M tokens |
| Output | ~$6 / M tokens |
Frontier models can run several times these rates, and that multiplier hits every token you send — including the context re-sent on every turn of an agent run. That is why the same task can be nearly free in Auto and expensive when forced onto a frontier model. For how per-turn context re-sends dominate a real session, see our breakdown of AI coding agent costs.
For most of what a developer does in a day, Auto is not a compromise — it is genuinely the right tool. The efficient models it routes to are more than capable of:
These tasks do not benefit from a frontier model's extra reasoning, so paying frontier rates for them is pure waste. Leaving Auto on for this 80% of your work keeps your credit pool untouched and available for the moments that actually earn it. If you use Cursor heavily, our guide to reducing Cursor costs covers this habit and several others.
Auto is the default, not a rule. There are tasks where a frontier model genuinely earns its cost, and overriding is the right call:
The discipline is to escalate deliberately: switch to Opus or GPT-5 for the hard part, get the result, then switch back to Auto. The mistake is pinning a frontier model as your permanent default and letting it handle typo fixes at 5-15x the cost. Override with intent, and undo the override when the hard part is done.
The whole strategy fits in three habits:
Model choice is only half of the cost equation — the other half is how many tokens each request carries. Even in Auto, a bloated prompt re-sent on every turn adds up. This is the layer Terse targets: it compresses the prompts you type into Cursor on-device and tracks the per-request cost with a light touch, so the token math is visible whether you are in Auto or on a frontier model. And if you want the fuller cost picture, our AI coding agent costs breakdown shows where the tokens really go.
Terse compresses the prompts you send into Cursor and tracks per-request token cost — on-device, zero latency, no API calls. Keep requests lean whether you are running Auto or a frontier model.
Terse for CursorAuto mode lets Cursor automatically choose a cost-efficient model for each request instead of pinning one frontier model. Usage in Auto is included at no extra cost and does not draw down your monthly credit pool, which makes it the sensible default for most day-to-day coding.
No. Requests handled in Auto mode are included in your Cursor Pro subscription and do not consume your roughly $20 monthly credit pool. Manually selecting a frontier model like Claude Opus or GPT-5 is what draws from that pool, and overages after it is spent bill at API rates.
Override Auto for genuinely hard reasoning, tricky refactors, or subtle bugs where a frontier model's extra capability earns its cost. For routine edits, boilerplate, small tests, and most day-to-day work, Auto is more than good enough and keeps your credit pool intact.
Once your monthly credit pool is exhausted, you can keep working in Auto mode at no extra cost, or continue on frontier models by paying overages at API rates. There is no automatic downgrade, so if you stay pinned to a frontier model you keep paying per token until the month resets.