Name: Mac mini (M6)
Brand: Apple

Running your own AI model used to mean owning hardware most people never encounter: server racks, GPU cards that cost more than laptops, subscriptions to remote compute. The M6 Mac mini, a small grey box the size of a thick book, now does the same work from a desk, for $899.

That price matters because it lands local AI — inference that runs without sending data to a remote server — inside the range of equipment many professionals already budget for. Apple‘s benchmarks show the M6 at 4.8 times the LLM prompt processing speed of the M4 model it replaces, measured in LM Studio. In practical terms: a large language model that previously needed specialised hardware runs on this one, entirely offline.

The M6 chip inside the machine carries a dual 16-core Neural Engine — two neural accelerators working simultaneously — which Apple says makes neural computation twice as fast as the previous generation’s. It pairs that engine with a 12-core CPU and a 12-core GPU backed by 170 GB/s of unified memory bandwidth. The standard configuration ships with 16 GB of unified memory, configurable to 32 GB.

The workloads it handles on-device include diffusion models for image generation, real-time photo style transfer, multi-step AI agents, and full LLM inference — all locally, all without a subscription. Apple Intelligence in macOS 27 exposes many of these through the operating system itself, while tools like LM Studio allow users to load any open model they want to run.

For context on how much has changed: Apple’s figures put the M6 at 13.5 times the LLM processing speed of the M1 Mac mini, a machine sold five years ago. The M1 could not run most local AI workloads in any practical sense. The M6 can.

A Mac mini with the M5 Pro chip is also available at $1,699 for users who need greater memory ceiling or higher multi-core performance. Pre-orders are open now; both models ship September 22.

The catch is real. A $899 machine still needs a monitor, keyboard, and mouse, which push the entry cost higher. And local AI inference is not simpler than cloud AI — it requires sourcing models, managing updates, and handling infrastructure that the cloud normally abstracts away. The people this makes most sense for right now are privacy-sensitive professionals who do not want data leaving their building, developers who want fast inference without per-token billing, and researchers who need a persistent local endpoint.

That group has been waiting for hardware priced like a tool rather than a server. With the M6, Apple is finally building for them.

Tags: Apple, Local AI, AI inference, Apple Silicon, M6 chip, Mac mini