a hybrid kernel

A hybrid kernel is the pragmatic middle ground between the two extremes. It is structured a bit like a microkernel - with the idea of separable subsystems and message passing in its design lineage - but for speed it keeps performance-critical services, like the filesystem and graphics, running inside kernel space rather than out in separate user-mode servers. In short: organized with microkernel ideas, but running mostly monolithically for the parts where the cost of crossing a boundary would hurt.

The clearest real examples are the Windows NT kernel and Apple's XNU. XNU (pronounced 'ex-en-you', the kernel of macOS and iOS) literally bolts the Mach microkernel together with a large chunk of BSD Unix code and runs them together in kernel mode - you get Mach's abstractions for messaging and memory, but BSD's filesystem and networking execute as fast in-kernel code, not as remote servers. Windows NT was designed with a microkernel-influenced layered structure, yet major subsystems were moved into kernel space (graphics famously moved in for performance) so that common operations do not pay message-passing costs.

The honest caveat is that 'hybrid kernel' is a loose, somewhat marketing-flavored label, and not everyone agrees it is a distinct category - critics (including some microkernel purists) argue that if the filesystem and drivers run with full privilege in kernel space, you essentially have a monolithic kernel that was merely organized using microkernel terminology. The useful takeaway is the spectrum, not the box: real kernels choose, subsystem by subsystem, how much isolation to trade for how much speed.