Hey, thanks for the report. This isn’t a bug. Sonnet 4.5 (especially the thinking variant) is counted as 2 requests. Reasoning models use extended thinking and generate extra hidden reasoning tokens, so they cost more compute and are billed with a multiplier.
Same situation with Opus 4.5. It’s also 2x requests.
Can you check the dashboard at cursor.com/dashboard?tab=usage ? You should see token details and the actual cost there. If anything looks off in the usage, send a screenshot and we’ll take a look.