I’ve noticed what feels like a more aggressive counting of usage for Auto+Composer recently. Still no visibility into the exact tokens consumed for these features, which makes it hard to track.
Does anyone else share this impression? In my last billing cycle, I was hammering Composer 1.5 and Auto extensively—whole days of coding without even glancing at usage metrics. Now, after just a few hours, I’m already at 1% utilization, compared to maybe 0.5% after a full day previously.
Curious if this is a backend change, a bug. Any thoughts?
Auto model selection: Auto dynamically picks the best model for each task. Auto pricing is fixed at $1,25 per 1M input tokens, $6,00 per 1M output tokens, and $0,25 per 1M cache reads, but the total tokens per task can vary depending on which model gets picked.
Composer 1.5 vs current models: If you used Composer 1.5 in the previous cycle and Auto is now choosing different models, token usage patterns can change a lot.
Dashboard breakdown: You can check a per-model token breakdown for each request at https://cursor.com/dashboard?tab=usage. This will show exactly which models are being used and how many tokens each request uses.
Can you share a screenshot of your usage dashboard? That’ll help us tell whether there’s a counting issue or just a change in model selection patterns.
However, it would be really helpful if the token usage for Composer 1.5 and Auto could also be displayed directly in the Billing tab. I don’t want to have to manually add up the tokens used by each Composer 1.5 and Auto.
