Hey. Auto doesn’t have a fixed token limit. Under the hood it works as a shared budget Auto plus Composer pool, separate from the API pool. The token count you see in usage history depends on which models the router picked this month, how well prompt caching worked, and those models’ per token prices.
So swings like 450 to 500M down to 300M month to month are normal and don’t mean the limit was cut. If routing went to more expensive models more often this month, or caching hit less, the same budget will produce fewer tokens.
More context on how the pool works: Models & Pricing | Cursor Docs and this thread Clarification on Cursor Pro vs Team Plan – Agent Usage and Token Allocation