Meta has open-sourced Muse Spark Open, an 800-billion-parameter model it calls the most capable US open-weight model yet. Running it locally requires roughly 1TB of unified memory, achievable by pairing two of Apple's upcoming M5 Ultra Mac Studios with 512GB each, or four 256GB units costing about $38,000. The weights are now downloadable from Hugging Face, and Meta says the model was built in collaboration with Apple to run on its new hardware.
engadget.com
· 2026-10-15
Meta launched Muse Spark Open, an 816-billion-parameter multimodal AI model it calls the most capable US open-weight model available, and said it can run on two Apple M5 Ultra Mac Studios using Apple's MLX framework. Apple published its own technical post on MLX and the new model, and executives from both companies publicly praised the collaboration, which Axios reported had been months in the making.
theverge.com
· 2026-10-15
Apple researchers detailed a system called Minions in an arXiv paper, which harnesses idle Apple Silicon Macs across the company's network to process AI inference workloads. The system splits jobs—embedding tasks and LLM text generation via Ollama—into subtasks that idle Macs pull and execute, cutting wait times from over 10 hours in GPU queues to minutes.
tomshardware.com
· 2026-10-14
A new paper describes Minions, a distributed system that aggregates spare GPU capacity from macOS devices running Apple Silicon chips to run large-scale machine learning inference workloads. The authors say the system handles task scheduling, fault tolerance, and resource utilization across these devices without requiring teams to buy additional GPU hardware.
machinelearning.apple.com
· 2026-10-13
A developer has released a free macOS app that converts Google's EmbeddingGemma 2 multimodal embedding model into a speech-to-text tool, transcribing an hour of audio in about 13 seconds on a base MacBook Air. The system keeps the original audio encoder untouched and instead trains a small 'reader' model to map audio slices into the text embedding space, matching them to the closest of roughly 229,000 English words, with a separate model adding punctuation. The app, EmbeddingGemma Transcribe, runs entirely on-device using the Mac's Neural Engine and currently has a word error rate of about 5%, compared with 2-3% for leading dedicated speech models.
eg2.haku.sh
· 2026-10-13
Apple's upcoming foldable iPhone Duo and the iPhone 18 Pro line will both use the new A20 Pro chip, built with an 'M-class-inspired' packaging design borrowed from Mac processors. To manage heat in the foldable's tight, hinge-divided chassis, Apple engineered a graphite thermal pathway that runs through the hinge itself to spread heat away from the chip. Apple silicon engineering VP Dan Murray and thermal design manager Banafsheh Barabadi described the design choices to accommodate one chip across two very different phone shapes.
pcmag.com
· 2026-10-13
Apple is preparing to launch a 14-inch MacBook Pro with its new M6 chip later this month, keeping the current design. The M6 adds a Dual 16-core Neural Engine for faster on-device AI processing, a 12-core CPU with a new three-tier core structure (super, performance, and efficiency cores), a 12-core GPU, and is built on a new 2nm manufacturing process that Apple says improves power efficiency.
9to5mac.com
· 2026-10-12
BetterWispr is a new Apache 2.0-licensed dictation tool for macOS 14 and later that requires no subscription, account or API key. It runs speech-to-text locally using downloadable models or Apple's on-device speech recognition, letting users dictate into any text field via a hold-to-record shortcut like Option+Space.
betterwispr.com
· 2026-10-11
Developer Skylar Taylor-Barrick built a project called gbos-vm that runs Google's Android-based Googlebook OS inside a virtual machine on Apple Silicon Macs, complete with hardware-accelerated graphics and a working Chrome browser. The tool bridges Android's Vulkan graphics calls to Apple's Metal system using POSIX shared memory and MoltenVK, letting the VM tap the Mac's GPU instead of relying on slower CPU rendering. It installs from a Dell Googlebook recovery image via a script and QEMU, requiring an Apple Silicon Mac with 16GB RAM, about 60GB storage, and developer tools like Xcode Command Line Tools and Homebrew.
techspot.com
· 2026-10-11
Google has introduced a new AI-powered note-taking application for Mac that operates entirely offline using on-device models. The app, named Google AI Edge Foresight, allows users to record, transcribe, and review meeting notes across various apps, supporting manual input and AI-generated summaries. It also enables document uploads to build a knowledge base for real-time question answering during meetings.
techcrunch.com
· 2026-10-08
Developers have created an open-source installer that allows users of Apple Silicon Macs to run Google's experimental desktop operating system within a virtual environment. This tool provides a way to test Googlebook OS without official support from Google or Apple.
androidauthority.com
· 2026-10-08
Google has released an experimental app called Google AI Edge Foresight that transcribes meetings and audio files entirely on-device, using its EmbeddingGemma 2 model. The free app turns shorthand notes into polished summaries and includes a built-in assistant that can answer questions about meetings or connected local files, with Google saying data never leaves the user's computer. It currently only works on Macs with Apple Silicon.
theverge.com
· 2026-10-08
GoKawiil