On August 25, 2026, Apple introduced the M6 chip in the new Mac mini and the M5 Ultra chip in the new Mac Studio, marking the company's first use of a 2-nanometer process and its first quad-die architecture in an M-series system on a chip. The M6 features a 12-core CPU, 12-core GPU with Neural Accelerators, a Dual 16-core Neural Engine, and up to 170 GB/s of unified memory bandwidth. The M5 Ultra combines two M5 Max dies via next-generation UltraFusion to deliver up to a 36-core CPU, up to an 80-core GPU, a 32-core Neural Engine, up to 512 GB of unified memory, and 1.2 TB/s of memory bandwidth — 50 percent more than the M3 Ultra.
Apple says the M6 is built for everyday users, students, developers, and AI hobbyists, while the M5 Ultra targets professionals running complex 3D rendering, visual effects, scientific analysis, and frontier AI models locally. Both chips are available in new hardware that can be pre-ordered starting today, with availability beginning September 22.
What's new
- M6 (Mac mini): 2 nm process; 12-core CPU (2 super, 4 performance, 6 efficiency); 12-core GPU with Neural Accelerators; Dual 16-core Neural Engine (up to 2× peak AI compute vs. previous generation); up to 32 GB unified memory; up to 170 GB/s memory bandwidth (10% over M5, 2.5× over M1).
- M5 Ultra (Mac Studio): Quad-die via UltraFusion (over 4.4 TB/s inter-die bandwidth, 6× connection density); up to 36-core CPU (12 super, 24 performance); up to 80-core GPU with Neural Accelerators; 32-core Neural Engine; up to 512 GB unified memory; 1.2 TB/s memory bandwidth (50% over M3 Ultra).
- Mac Studio updates: Wi‑Fi 7, Bluetooth 6, Thunderbolt 5 (supports clustering multiple Mac Studios for up to 3× faster distributed AI inference via RDMA); up to 2× faster storage; up to 1.8× faster graphics; up to 1.3× faster CPU vs. prior generation.
- Software: macOS 27 with next-generation Apple Intelligence including Siri AI; Core AI framework for building, running, and deploying AI models on Apple silicon; MLX open-source framework; Xcode with on-device coding agents.
Why it matters
The M6 brings Apple's first 2 nm silicon to a mainstream desktop, raising the baseline for on-device AI with a Dual Neural Engine and higher memory bandwidth in a compact Mac mini. The M5 Ultra's quad-die design and 512 GB unified memory pool let researchers and developers run models with hundreds of billions of parameters entirely on device — no cloud tokens, no recurring cloud costs. Thunderbolt 5 clustering extends that capability further, turning multiple Mac Studios into a shared-memory AI compute cluster. For pro video, the M5 Ultra's media engine adds four ProRes encode/decode engines and hardware AV1 decode.
Our take
Apple is using the M6 to democratize local AI at the entry level and the M5 Ultra to own the high-end workstation AI niche where unified memory capacity still beats discrete GPU VRAM per dollar. The real test will be whether the software ecosystem — Core AI, MLX, and third-party frameworks — can keep pace with the hardware's token throughput claims when developers move beyond Apple's curated demos.
