Apple quietly introduced a native container runtime in macOS 26—and it doesn’t look like Docker/Podman under the hood. Instead of running all containers inside a single Linux host VM, Apple’s CLI (“container”) spins up a lightweight VM per container via Virtualization.framework. That means each container gets its own kernel, IP, ext4 block storage, and explicit CPU/memory limits. On M3+ Macs, you can even expose nested virtualization (I put a VM in your Container!) It’s OCI‑compatible (your existing Docker/Podman/Kubernetes images work), and Rosetta 2 even lets you run amd64 images on Apple Silicon.
I benchmarked Apple’s runtime (v0.5.0) against Docker/Colima on an M1 Pro (32 GB, macOS 26.0.1). I measured image pulls, cold/warm starts, lifecycle ops, parallel starts, file churn, plus stress‑ng, fio, and 7zip.
A few takeaways:
Startup: sub‑second starts as advertised; “container system start” returns instantly (no host VM spin‑up).
CPU/Memory: competitive or slightly favorable to Apple in stress‑ng and 7zip; memory tests consistently leaned Apple.
I/O: Fio (the flexible I/O tester!) flipped the story—Docker performed substantially better on randomized reads and mixed RW.
Clickthrough to the post to find exact commands, scripts, and full outputs, plus charts comparing Apple vs Docker/Colima.
Author here -
Apple quietly introduced a native container runtime in macOS 26—and it doesn’t look like Docker/Podman under the hood. Instead of running all containers inside a single Linux host VM, Apple’s CLI (“container”) spins up a lightweight VM per container via Virtualization.framework. That means each container gets its own kernel, IP, ext4 block storage, and explicit CPU/memory limits. On M3+ Macs, you can even expose nested virtualization (I put a VM in your Container!) It’s OCI‑compatible (your existing Docker/Podman/Kubernetes images work), and Rosetta 2 even lets you run amd64 images on Apple Silicon.
I benchmarked Apple’s runtime (v0.5.0) against Docker/Colima on an M1 Pro (32 GB, macOS 26.0.1). I measured image pulls, cold/warm starts, lifecycle ops, parallel starts, file churn, plus stress‑ng, fio, and 7zip.
A few takeaways: Startup: sub‑second starts as advertised; “container system start” returns instantly (no host VM spin‑up).
CPU/Memory: competitive or slightly favorable to Apple in stress‑ng and 7zip; memory tests consistently leaned Apple.
I/O: Fio (the flexible I/O tester!) flipped the story—Docker performed substantially better on randomized reads and mixed RW.
Clickthrough to the post to find exact commands, scripts, and full outputs, plus charts comparing Apple vs Docker/Colima.