Skip to content
What are you looking for?. Open searchSR
Machines & local modelsstrong signalverified

Apple claims 4.3x faster AI on Mac Studio, and the 512GB configuration arrives in late October

Mac Studio with M5 Max and the all-new M5 Ultra reaches stores on September 22, 2026, and Apple puts its AI performance at up to 4.3x the previous generation, measured in July 2026 against a machine with M3 Ultra. The configuration that decides how large a local model can be, 512GB of unified memory at up to 1.2TB/s, sits in a footnote: it comes in late October. On the software side the new Siri AI ships in beta, in English to start, and will not be available in the EU on iOS, iPadOS or watchOS at launch. Apple Intelligence features that run on server-side models also have daily usage limits, and Apple says wider access to them will be sold for a fee later.

By Redakcija WebAiRadarPublished 3 min readwritten by a model
Image: Apple

Source

Get ready to experience iPhone 18 Pro, the next generation of Apple Watch, and AirPods 5

Apple Newsroom · Original published September 11, 2026

Apple opened pre-orders and set the dates on September 11, 2026. Pre-orders for iPhone 18 Pro and iPhone 18 Pro Max begin on September 12, 2026, and those phones reach stores worldwide on September 18, 2026, together with Apple Watch Series 12, Apple Watch Ultra 4 and AirPods 5. Mac mini and Mac Studio, the part of this launch that matters to anyone running a model on their own desk, reach stores on September 22, 2026.

What the speed figures are measured against

Apple puts Mac Studio with M5 Ultra at up to 4.3x faster AI performance, up to 2x faster storage, up to 1.8x faster graphics and up to 1.3x faster CPU speed. The footnotes say what the comparison is: a previous-generation Mac Studio with M3 Ultra, a 32-core CPU, an 80-core GPU, 512GB of unified memory and an 8TB SSD, tested by Apple in July 2026.

Mac mini gets the same treatment, at 40 percent faster CPU performance against a previous-generation Mac mini with M4, a 10-core CPU, a 10-core GPU, 32GB of unified memory and a 2TB SSD. No benchmark is named for either machine and no absolute figure is given, so these are the vendor's own ratios rather than a result anyone can reproduce from the page.

The memory line, and the date underneath it

For running a model locally, the figure that decides what fits is memory. Mac Studio goes up to 512GB of unified memory and up to 1.2TB/s of memory bandwidth, with Thunderbolt 5, Wi-Fi 7 and Bluetooth 6. Mac mini comes with the all-new M6 or with M5 Pro, and Apple describes it as a machine for running AI models on device.

A footnote carries the catch. Mac Studio with 512GB of unified memory is coming in late October, not on September 22, 2026 with the rest of the line. The page gives no starting prices, though its leasing examples use a Mac Studio purchase price of $2,499, an iPhone 18 Pro 256GB at $1,199 and an Apple Watch Series 12 GPS 42mm at $399.

Siri AI arrives with a border and a meter

The new Siri is rolling out in beta with iOS 27, macOS, iPadOS 27 and watchOS 27. It requires an Apple Intelligence-enabled device set to a supported language, it is available in English to start, and it is not available to users under 13.

Two limits appear in the small print, and both are terms of use rather than technical notes. Apple says Siri AI will not be initially available in the EU on iOS, iPadOS or watchOS, and that features relying on Siri AI are unavailable there as well. It adds that it is working to find a path forward that preserves users' privacy and security.

  • Certain Apple Intelligence features that rely on server-side models are subject to daily usage limits.
  • Those features include Siri AI, the intelligent photo editing tools, Image Playground and AFM 3 Cloud models in Shortcuts.
  • Apple says the limits vary by feature, request complexity, system demand, system policies and other factors, and it publishes no number.
  • Expanded access to those features will be available for a fee in the future.
„Mac Studio with 512GB of unified memory is coming in late October“
Apple, iPhone 18 Pro availability announcement
TagsLocal inferencelocal modelsHardwareApple siliconprivacy

Related

LLAMA.CPP B115567.63 GBBonsai 2 27B without the fork
Machines & local modelsmedium signal

llama.cpp now runs PrismML's ternary Bonsai 2 27B without the company's fork

Build b11556 adds the weight rotation step that Bonsai 2 27B was missing in stock llama.cpp. The 7.63 GB Q2_0 file now gives correct output there, while PrismML's smaller PQ2_0 and PTQ1_0 packs still require the company's own build.

llama.cppverified

LLAMA.CPP0.6.0GLM-5.3-Flash and a server API for decision mo
Machines & local modelsmedium signal

llama.cpp 0.6.0 adds GLM-5.3-Flash support and a server API for decision models

The llama.cpp project released version 0.6.0 on October 5, 2026. It adds support for GLM-5.3-Flash, a model for text and images with 320 billion parameters, and a new server endpoint for decision models. Two changes to speculative decoding come with measurements from their authors, one on Nvidia's DGX Spark and one on Apple GPUs. The release also changes the C API and raises the version of the saved session format, so programs built on the library need a check before you upgrade.

llama.cppverified