Schemas and validation, middleware hooks (pre and post), population for relationships, instance and static methods, and a big community. You write less boilerplate and catch bad data at the schema level.

Slower than the raw driver due to hydration and validation. More magic that you sometimes fight. Heavier learning curve for advanced features like discriminators and complex middleware.

For high-throughput writes where Mongoose's overhead matters, highly variable schemas (logs, events), complex aggregations where you want full control, or migrations where you want minimal abstraction.