Apple has announced new 2026 editions of the Mac mini and Mac Studio, surprisingly before the autumn iPhone event rather than after it.
The Mac mini gets the brand new M6 chip, Apple’s first built on a 2nm process, with an M5 Pro option above it. The Mac Studio moves to M5 Max and a new M5 Ultra, the first quad-die chip Apple has made, with up to 512GB of unified memory.
Pre-orders open today, 25 August, in 30 countries and regions, and both machines go on sale from 22 September. In the UAE, the Mac mini starts at AED 3,799 and the Mac Studio at AED 10,499.
Apple briefed tbreak on the new hardware ahead of the announcement, and the message throughout was about one thing: running AI locally. Apple’s Mac business grew almost 30% last quarter, and the company credits a good part of that to people buying Mac mini and Mac Studio units to run models and agents on their own desks rather than in someone else’s cloud. Both new machines are pitched primarily at that crowd.
M6: Apple’s first 2nm chip
The M6 is built on a 2nm process, which Apple says packs greater transistor density into a smaller die and gives it enough headroom to upgrade every compute block on the chip. It is a two-generation jump for the Mac mini, which was last updated with M4, but Apple’s own comparisons are mostly against the M5 in the current MacBook Pro, so it is worth reading the numbers with that in mind.
The CPU is a 12-core design, two more cores than M5 and the first time a mainstream M-series chip has gone beyond ten. It is also the first Apple silicon to combine all three of Apple’s CPU core types: two super cores that Apple claims deliver the fastest single-threaded performance in the industry, four performance cores for multi-threaded work, and six efficiency cores for background tasks. Apple puts multi-threaded performance at up to 1.2x M5 and 2.4x M1, and specifically calls out agentic AI workloads, code compilation and file indexing as the sort of thing that benefits.
The GPU is also 12 cores, again two more than M5, and it brings Apple’s Neural Accelerators to the Mac mini for the first time. That is the big number for local LLM work: nearly 30% more peak GPU compute for AI than M5, more than 8x M1, and up to 4x the M4 the mini is replacing. The GPU also picks up the M5-generation shader core, Dynamic Caching and hardware ray tracing, with 50% higher geometry rates for complex scenes. Graphics performance overall is up to twice that of the M4 mini.
Then there is a dual 16-core Neural Engine, which Apple says doubles peak compute for power-efficient on-device models. System frameworks can use both engines at once, so apps get the benefit without being rewritten. Unified memory bandwidth rises to 170GB/s, which is 10% more than M5, 2.5x M1 and roughly 40% up on M4, with capacity topping out at 32GB.
The new Mac mini: M6 or M5 Pro
The Mac mini itself keeps the compact design and the same thermal architecture as the current model, but almost everything inside has changed. The base configuration pairs the M6 (12-core CPU, 12-core GPU) with 16GB of unified memory and a 256GB SSD for AED 3,799, and memory is configurable up to 32GB. Apple frames the M6 mini as the always-on desk-side box that runs agents next to your MacBook, and it name-checks agentic coding in Xcode, games like Control, upscaling in Topaz Photo, and agent frameworks such as OpenClaw and Codex.
Apple’s own benchmarks for the M6 mini, all against the M4 model: up to 4.8x faster LLM prompt processing in LM Studio (13.5x against an M1 mini), up to 1.5x faster spreadsheet calculations in Excel, and up to 2x faster gaming with ray tracing on in Cyberpunk 2077: Ultimate Edition.
For heavier work, the M5 Pro option arrives with up to an 18-core CPU and up to a 20-core GPU, Neural Accelerators in every GPU core and Thunderbolt 5. It starts at AED 7,299 in the UAE with a 15-core CPU, 16-core GPU, 24GB of unified memory and a 512GB SSD. This is the configuration to look at if you want to run larger models locally: it supports up to 64GB of unified memory at 307GB/s of bandwidth, and Apple claims up to 4x faster LLM prompt processing than the M4 Pro mini, along with 1.4x faster Blender rendering and 1.5x faster image processing in Affinity.
On connectivity, the front of both minis has two USB-C ports (USB 3) and a headphone jack with high-impedance headphone support. Round the back, the M6 model has three Thunderbolt 4 ports and the M5 Pro has three Thunderbolt 5, plus HDMI and Ethernet. Wired networking moves to 2.5Gb Ethernet as standard, with 10Gb still an upgrade, and both get Apple’s N1 wireless chip with Wi-Fi 7 and Bluetooth 6. Storage is up to twice as fast as before. Thunderbolt 5 on the M5 Pro also means multiple minis can be clustered to run larger models on device, and there is new genlock support over USB-C for syncing a display with a camera such as the iPhone 17 Pro. If you are coming from an M1 Mac mini, Apple’s figure is over 20x faster for workloads like AI inference.
Mac Studio: M5 Max and the quad-die M5 Ultra
The Mac Studio is getting what Apple describes as its biggest performance jump since the line was introduced: up to 4.3x faster AI performance, up to 1.8x faster graphics, up to 1.3x faster CPU and twice the storage speed over the M3 Ultra model it replaces.
The entry chip is now the M5 Max, already seen in the MacBook Pro. It has an 18-core CPU made up of 6 super cores and 12 performance cores, an up-to-40-core GPU with Neural Accelerators in every core that is up to 50% faster than the previous generation, up to 128GB of unified memory and 614GB/s of bandwidth. Apple puts its AI performance at up to 3.9x the prior Studio, which shows up as 3.9x faster LLM prompt processing in LM Studio and 3.5x faster text-to-image generation against the M4 Max, and 3x faster Magic Mask in DaVinci Resolve Studio. The base UAE configuration, at AED 10,499, uses an M5 Max with the 18-core CPU and a 32-core GPU, 36GB of unified memory and a 512GB SSD, with the 40-core GPU sitting above it.
Then there is the M5 Ultra, a two-generation jump from the M3 Ultra. It uses a new version of Apple’s UltraFusion interconnect to join two M5 Max chips, and because each M5 Max is itself a dual-die design, the M5 Ultra is a quad-die chip, a first for Apple silicon. The new interconnect carries over 4.4TB/s between dies with more than six times the connection density of before, and software still sees the whole thing as a single processor with a single pool of unified memory.
The full M5 Ultra has a 36-core CPU (12 super cores and 24 performance cores) that is up to 1.25x faster single-threaded and 1.3x faster multi-threaded than M3 Ultra, and an 80-core GPU that brings Neural Accelerators to the Ultra tier for the first time, giving up to 4.3x the peak AI compute of the M3 Ultra Mac Studio and 9.8x an M1 Ultra.
Graphics are up to 1.8x faster, with second-generation Dynamic Caching, mesh shading and third-generation ray tracing. There is a 32-core Neural Engine, and a Media Engine with twice the encode and decode blocks of M5 Max, four ProRes engines and AV1 decode, enough to play 33 streams of 8K ProRes 422 at 30fps simultaneously. Memory bandwidth is up 50% to 1.2TB/s, with capacity up to 512GB. Apple’s own numbers against M3 Ultra: 4x faster LLM prompt processing in LM Studio, 4.3x faster text-to-image, 3.3x faster CopyCat training in Nuke and 1.7x faster rendering in Redshift.
The entry M5 Ultra Studio, at AED 23,999, ships with a 30-core CPU, 64-core GPU, 96GB of unified memory and a 1TB SSD. Both Studio models move to a new PCIe SSD architecture that is up to twice as fast as before, useful for loading large models and video projects, along with the N1 chip for Wi-Fi 7 and Bluetooth 6, both new to the Studio. Thunderbolt 5 delivers up to 120Gb/s for external storage, PCIe expansion chassis and hubs, genlock over USB-C is supported here too, and the Studio can drive up to eight displays, or four Studio Display XDRs at full 5K and 120Hz. Against an M1 Mac Studio, Apple claims up to 15x the performance in workloads such as AI model training.
Clustering: four Mac Studios from a wall socket
The part of the announcement aimed at the most ambitious local-AI users is clustering. Thunderbolt 5 and RDMA (remote direct memory access) let multiple Mac Studios pool their memory into one shared space, and Apple is now openly positioning a four-node cluster as a way to load the largest frontier-class open-weight models, including those with over a trillion parameters, with up to 3x the inference performance of a single Mac Studio. What’s incredible is that the whole cluster runs off a standard wall outlet thanks to the efficiency of Apple silicon, and that once the hardware is paid for there are no per-token costs to think about.
Software: macOS 27 Golden Gate and Core AI
Both machines ship with macOS 27, codenamed Golden Gate, which brings the new Siri along with a wider set of Apple Intelligence features. Beyond AI, Apple is promising a snappier Mac overall: faster AirDrop and network file browsing, quicker Safari start pages, a rebuilt search index behind Spotlight, Photos and Mail, and refinements to Liquid Glass including uniform toolbars, edge-to-edge sidebars and more detailed app icons. Maps, Notes and Podcasts get updates too. macOS 27 is in public beta now and arrives as a free update this autumn.
On the developer side, the notable addition is Core AI, a new framework for building, running and deploying AI models on Apple silicon that is designed around unified memory and spreads work across CPU, GPU and Neural Engine. Together with MLX, Core ML, Metal and Xcode, it is how tools like LM Studio, Draw Things, Cinema 4D and Final Cut Pro get their speed-ups on the new machines, and developers can use Apple’s Foundation Models or bring their own.
UAE pricing, pre-orders and availability
Here is how the four base configurations line up in the UAE. Every model can be configured further, and the M5 Pro mini, M5 Max and M5 Ultra all have fuller-core versions of their chips available above the entry spec.
| Mac mini M6 | Mac mini M5 Pro | Mac Studio M5 Max | Mac Studio M5 Ultra | |
|---|---|---|---|---|
| CPU | 12-core | 15-core (up to 18) | 18-core | 30-core (up to 36) |
| GPU | 12-core | 16-core (up to 20) | 32-core (up to 40) | 64-core (up to 80) |
| Unified memory | 16GB (up to 32GB) | 24GB (up to 64GB) | 36GB (up to 128GB) | 96GB (up to 512GB) |
| Memory bandwidth | 170GB/s | Up to 307GB/s | Up to 614GB/s | Up to 1.2TB/s |
| Storage (base) | 256GB | 512GB | 512GB | 1TB |
| Thunderbolt | 3x Thunderbolt 4 | 3x Thunderbolt 5 | Thunderbolt 5 | Thunderbolt 5 |
| UAE price | AED 3,799 | AED 7,299 | AED 10,499 | AED 23,999 |
| UAE education price | AED 3,399 | AED 6,899 | AED 9,699 | AED 22,299 |
| US price | US$899 | US$1,699 | US$2,499 | US$5,499 |
Pre-orders open today, 25 August, on apple.com and in the Apple Store app, and both machines reach customers, Apple Stores and authorised resellers from 22 September. The 512GB memory option for Mac Studio arrives later, in late October.
Why the early launch?
Apple usually refreshes Macs in October, after the iPhone. I asked why the desktops are landing early this year- Apple’s explanation was demand: people are waiting for these two machines in particular, and the company would rather get them on sale than hold them for a later event. Considering the RTX Spark is landing soon as well, I wouldn’t be surprised if Apple wanted to get ahead of the Microsoft/Nvidia release.
It was also asked about the memory supply constraints flagged on its last earnings call, and offered no assurances beyond saying it will build as many as it can. Given how much memory the top configurations carry, and the late-October date for the 512GB option, pre-ordering early looks sensible if you want a high-memory Studio.
One more question that came up was whether the surge in Mac mini sales for agent frameworks like OpenClaw shaped these products. The answer is that the hardware was already well into development when that wave hit; the mini simply turned out to be the right shape for it. Apple’s line is that it builds a good general-purpose computer first and the AI performance follows, and the M6 spec sheet, upgraded across CPU, GPU, Neural Engine and memory, does back that up.
For UAE buyers, the split is fairly clear. The AED 3,799 M6 mini is the entry point for running agents and smaller models on the desk; the AED 7,299 M5 Pro mini, with its 24GB base and 64GB ceiling, is where larger local models start to fit; and anything beyond that is Mac Studio territory, where 96GB and above turns the question from whether a model runs into how fast.
