Apple is asking buyers to pay more for its smallest desktop while arguing that the compact machine has become something bigger: a building block for local AI.

The shift began in December, when macOS 26.2 enabled low-latency Thunderbolt 5 links for distributed inference using Apple’s MLX framework. That capability helped hobbyists, researchers and developers chain together Mac minis and Mac Studios to run larger local language models—an alternative to costly cloud tokens and Nvidia-heavy systems. Apple’s own release notes described the feature as supporting “distributed AI inference using MLX.”¹

Today’s refresh turns that emerging behavior into the sales pitch. The new Mac mini retains its 2024-sized enclosure but moves to an M6 base model and an M5 Pro option. The M6 brings a 12-core CPU, 12-core GPU and 170GBps of memory bandwidth; Apple says its CPU is up to 40% faster than the M4 generation. Both models gain faster storage, 2.5Gb Ethernet as standard—with optional 10Gb—and Wi-Fi 7 and Bluetooth 6 through Apple’s N1 chip.²

The two accounts converge on the central bet: more bandwidth and networking make the Mini a more credible local-AI workstation, not merely a cheaper Mac. But they frame the machine differently. One stresses Apple’s broader desktop strategy, including the Mac Studio and its ultra-high-memory configurations for production inference. The other focuses on the Mini’s more practical upgrades, saying the M5 Pro’s GPU neural accelerators should enable “heavier AI workloads,” including larger diffusion models and media upscaling.²

There is a catch. The M6 Mini starts at $899 and the M5 Pro version at $1,699—each $100 above the prior generation’s entry price. Apple is effectively testing whether developers squeezed by cloud-AI costs will see a premium Mini as a saving. Preorders begin now; machines are due to ship September 22 with macOS 27 Golden Gate.¹