TL;DR: Apple announced refreshed Mac mini and Mac Studio desktops on August 25, 2026, built around new M6 and M5 Ultra chips to run large AI models locally and in Thunderbolt 5 clusters.

Apple refreshes two desktops for on-device AI

Apple introduced new versions of the Mac mini and Mac Studio on August 25, 2026, positioning both systems for local AI inference and software development, according to Ars Technica. The company presented the update as a specifications bump without major new features for either enclosure, but framed the chips inside as the central change.

The release adds two processors to the lineup. The M6 is described as the first 2nm chip in Apple's M-series for Macs, while the M5 Ultra is described as the most powerful chip in the lineup for most tasks, especially AI workloads, according to Ars Technica. Configurations will also include M5 Pro in the Mac mini and M5 Max in the Mac Studio.

Apple said pre-orders for both computers started that same day, with availability beginning September 22. The 512GB memory configuration of the Studio was scheduled to follow later, with shipment in late October, according to Ars Technica. Apple said both systems will ship with macOS 27 Golden Gate.

Why local inference shaped the announcement

The Mac mini and Mac Studio occupy different places in Apple's desktop range, but both have become popular for local AI inference and development because of unified memory architecture and fast processors on their systems-on-a-chip, according to Ars Technica.

That interest accelerated after macOS 26.2 shipped last December. Apple's release notes for 26.2 said the update enabled low-latency communication between Thunderbolt 5 hosts for uses including distributed AI inference using MLX, according to Ars Technica. Thunderbolt 5 is a fast wired connection, and MLX is an open-source array framework intended to help machine learning workflows use M-series unified memory.

Since that software change, hobbyists, professional developers and researchers have connected multiple Mac minis or Mac Studios to run inference on local large language models larger than what could fit on one mass-market machine, offering an alternative to specialized Nvidia GPU hardware, according to Ars Technica. CNBC reported on August 25, 2026 that the new Mac Mini and Mac Studio models include updated chips and improvements aimed at artificial intelligence applications.

M6 brings new core layout to Mac mini

The M6 is a next-generation system-on-a-chip for a wide range of consumer uses, with a 12-core CPU containing two Apple-designated super cores, four performance cores and six efficiency cores, according to Ars Technica. Apple is presenting it as the first Apple SoC to use all three core types together.

Apple claims up to 40 percent faster multi-threaded CPU performance compared with the M4 from two generations earlier, though verifiable third-party benchmarks were not yet available at announcement time, according to Ars Technica. The chip also includes a 12-core GPU, two cores more than its immediate predecessors, alongside faster unified memory.

Apple's Mac mini announcement says the M6 system delivers up to 4x faster AI performance, 2x faster storage and graphics, and 40 percent faster CPU performance. The company lists a 12-core CPU with two additional cores versus the prior design, a 12-core GPU with two additional cores, Neural Accelerators in each GPU core for the first time on Mac mini, and a Dual 16-core Neural Engine with up to 2x faster performance than the previous generation, according to Apple. Standard unified memory starts at 16GB and is configurable to 32GB, with bandwidth up to 170GB/s.

For application examples, Apple cites up to 13.5x faster large language model prompt processing in LM Studio compared with Mac mini with M1, and up to 4.8x faster than M4. It also cites up to 2.3x faster spreadsheet calculations in Microsoft Excel compared with M1 and up to 1.5x faster than M4, plus up to 2x faster gaming with ray tracing in Cyberpunk 2077: Ultimate Edition compared with Mac mini with M4, according to Apple.

M5 Ultra targets very large models in Mac Studio

Memory capacity on the M6 Mac mini remains limited to 32GB, which is where the Mac Studio option enters the story, according to Ars Technica. The updated Mac Studio will offer M5 Ultra, with maximum capacity of 512GB for buyers selecting that configuration.

In many respects the processor combines two M5 Max dies side by side on one SoC, producing 36 CPU cores, including 12 super cores and 24 performance cores, plus 80 GPU cores, according to Ars Technica. Apple claims up to 1.2TB/s of unified memory bandwidth for that design.

Apple describes Mac Studio with M5 Max as using an 18-core CPU, a GPU up to 40 cores with Neural Accelerators built into each core, and up to 128GB of unified memory for complex professional and AI workloads, according to Apple. With M5 Ultra, the system scales to a 36-core CPU, up to an 80-core GPU and 512GB of unified memory, allowing users to run large language models entirely on the device.

The company claims up to 4.3x faster AI performance overall, up to 2x faster storage, up to 1.8x faster graphics and up to 1.3x faster CPU speed on the new Mac Studio, alongside higher memory bandwidth, according to Apple. It attributes faster matrix multiplication to Neural Accelerators in each GPU core.

Apple further breaks out AI compute gains by chip generation. M5 Max is listed at up to 3.9x faster AI performance than the prior generation for on-device work and faster prompt processing, according to Apple. M5 Ultra is listed at up to 4.3x the peak AI compute performance of M3 Ultra and 9.8x more than M1 Ultra. The company says bandwidth of 1.2TB/s is 50 percent higher than before, allowing large models to run on device with privacy and without token metering or cloud-cost concerns, according to Apple.

Clusters link multiple Studios over Thunderbolt 5

A major emphasis is scaling beyond one machine. Apple says multiple Mac Studio systems can be clustered using built-in Thunderbolt 5 and RDMA, or remote direct memory access, creating a shared memory pool across systems for loading large frontier-class open-weight models, according to Apple.

A cluster of four Mac Studio systems is rated at up to 3x faster AI inference than a single system, according to Apple. Ars Technica similarly notes that users have been daisy-chaining machines for distributed inference after the Thunderbolt 5 low-latency support arrived in macOS 26.2.

Apple also points to software frameworks in macOS for this workflow. Core AI is described as a new framework for building, running and deploying AI models on Apple silicon, optimized for unified memory, CPU, GPU and Neural Engine and able to deploy full-scale local large language models and custom models in apps, according to Apple. MLX, Apple's open-source machine learning framework optimized for Apple silicon, is described as a way to run, train and fine-tune models efficiently on Mac.

The company adds that developers can get faster build performance in Xcode and use on-device coding agents on the new Mac Studio, plus up to 4.3x faster text-to-image performance in apps compared with the previous generation, according to Apple.

Context: cloud costs and open-weight models

Ars Technica links the hardware direction to changing developer workflows around AI coding agents. Many developers have incorporated frontier large language models and harnesses such as Claude Code or Codex more heavily, but operating those cloud models carries steep token costs that raise questions about long-term practicality, according to Ars Technica.

Open-weight models that run locally, including recent Qwen and DeepSeek models, are presented as able to handle many of the same tasks without token charges because they are smaller and use the user's own local energy and compute, according to Ars Technica. At the same time, typical consumer hardware such as an average-spec MacBook Pro remains limited in the size of models it can run, which is why some users have turned to chains of Mac minis or Mac Studios.

Apple makes a similar argument in its own materials, saying Mac Studio lets users run massive models entirely on device without counting tokens or worrying about rising cloud costs, according to Apple. The Mac mini announcement describes the small desktop as suitable for productivity, creative work, running AI models on device and creating AI agents that automate daily tasks, according to Apple.

Connectivity, storage and Ethernet changes

Both refreshed machines include Apple's N1 chip, which supports Wi-Fi 7 and Bluetooth 6, according to Ars Technica. Apple says Wi-Fi 7 and Bluetooth 6 are coming to Mac Studio for the first time, with Thunderbolt 5 completing its connectivity for external storage, PCIe expansion chassis and hubs, according to Apple.

Apple claims storage up to twice as fast, at 15GB/s, according to Ars Technica. The Mac mini ships with 2.5Gb Ethernet as standard, with an option to upgrade to 10Gb, according to Ars Technica. Apple's mini announcement likewise lists upgraded 2.5Gb Ethernet with a 10Gb option on both Mac mini models, according to Apple.

For the higher-end mini, Mac mini with M5 Pro features up to an 18-core CPU and up to a 20-core GPU with an enhanced shader core and third-generation ray tracing for 3D design, visual effects and game development, according to Apple. It supports up to 64GB of unified memory with 307GB/s of memory bandwidth for larger local AI models, complex 3D scenes, ProRes RAW editing and large custom research datasets, according to Apple. The company cites up to 8.5x faster LLM prompt processing in LM Studio compared with Mac mini with M2 Pro, and up to 4x faster than M4.

Prices and configurations at launch

Pricing starts at $899 for Mac mini with M6 and 16GB of memory, while configurations with M5 Pro start at $1,699, according to Ars Technica. Mac Studio with M5 Max starts at $2,499, and M5 Ultra configurations start at $5,499, with higher costs depending on selected memory, storage and other options, according to Ars Technica.

Apple's hardware chief Johny Srouji said Mac Studio is intended as a desktop for on-device AI and demanding professional workflows in a compact desk-side design, adding that M5 Max and M5 Ultra bring large performance gains for professional work and AI inference with frontier-class models through GPU-integrated Neural Accelerators and high-bandwidth unified memory, according to Apple. For Mac mini, Srouji said the M6 delivers a new level of AI performance through a more powerful CPU and graphics, GPU Neural Accelerators and higher memory bandwidth, while M5 Pro provides an ultracompact option for complex professional workflows, according to Apple.

Sources referenced