Hi, I’m currently on the Cursor Pro Plan (Individual), and I have a question about how Auto mode works as of August 2026.
It appears that Auto usage is now counted toward my Cursor Models usage pool, but I’m not able to find clear information about the exact pricing or consumption rate.
I noticed that Auto has been reworked for Team plans, but for Individual developers, there doesn’t seem to be much detailed documentation beyond the “Auto cost” section, which appears relatively expensive. How exactly is Auto usage calculated compared with selecting a specific model directly?
Since Auto does not seem to provide a significant cost advantage over manually selecting models (for example, Grok, which is already a frontier model and sharing the same usage pool), I’m wondering whether Auto mode is still the best value option.
For current Pro users, do you still recommend using Auto, or is it generally better to manually select models depending on the task?
On Pro Individual, Auto is currently the classic Auto, which in the new terms is called Auto Cost. The key point is pricing is a flat per token rate, no matter which model the router picks in the background. If Auto ends up using a frontier model under the hood, like Grok 4.5, you still get charged the fixed Auto rate, not that model’s rate. That leads to two things: 1 Auto is often cheaper for many tasks than manually picking the same model, and 2 Auto Cost gives you a predictable spend cap because the rate is fixed.
The new Auto rework with three modes Cost, Balance, Intelligence is Cursor Router, and according to the team it’s currently only available on Teams and Enterprise. It’s not on individual plans yet. Staff confirmed this in the announcement Introducing Cursor Router see Colin’s replies in posts 14 and 17. So the Balance and Intelligence logic and billing at the actually selected model rate does not apply to your plan yet.
About the pool: Auto, Grok 4.5, and Composer 2.5 all draw from the same first party Cursor Models pool. That’s by design, not a change specific to Auto.
In practice:
Auto Cost: when you want the best per token value inside the pool and a predictable cap from the fixed rate, and you don’t care which exact model runs.
Manual model selection: when you need the specific behavior of a specific model.
When Router comes to individual plans, it’s worth double checking billing: Cost should keep the current flat rate logic bundled Auto pricing, while Balance and Intelligence will be charged at the rate of the model that actually gets picked. Let me know if anything’s still unclear.
thank you for the reply. Maybe i didn’t explain myself clear enough. the cost of AUTO is very similar to Grok 4.5 , why shouldn’t I manually select Grok by default? Right now I can’t justify the logic. Maybe a pricing reconsideration in the future? Or maybe a little bit more on the transparency of this classic AUTO?
Good clarifying question, I’ll focus on that specific part.
First, an important terminology note. Auto has three modes: Cost, Balance, and Intelligence. The flat pricing logic I’m talking about below applies specifically to Cost mode. On Pro Individual today, Auto works like classic Auto, meaning Auto Cost, so for your plan this is the relevant option. Balance and Intelligence are billed at the rate of whatever model the router picks, but they’re currently part of Cursor Router and aren’t available on individual plans.
Now to the point. The reason isn’t the nominal output rate, since Auto Cost and Grok 4.5 are indeed close there. It’s the traffic mix. In agent sessions, most tokens typically come from cache reads, not fresh input or output. Auto Cost is a fixed rate regardless of which model the router uses under the hood, and its cache read rate is about half of Grok 4.5 ($0.25/1M vs $0.50/1M). So even with similar output pricing, the real total session cost on Auto Cost often ends up cheaper than manually selecting Grok, especially in long chats with lots of context.
Two practical notes:
Auto often routes to Grok 4.5 under the hood anyway, but you’re charged the flat Auto Cost rate, not the model’s own rate.
Flat rate means predictability. The per token price doesn’t change depending on which model the router picked.
So picking Grok manually makes sense when you specifically need that model’s behavior or you want full determinism in model choice. If you don’t need that, Auto Cost is usually cheaper in practice.
For the exact current rates (input, cache write, cache read, output), it’s best to double check the official pages so you’re not relying on outdated numbers: Cursor · Pricing and Models & Pricing | Cursor Docs. Compare the cache read row for Auto Cost vs Grok. That difference is the most obvious.
On the transparency of classic Auto, fair point. I’ll pass it to the team as feedback. Let me know if after checking the cache read rates it still doesn’t add up.