Apple just refreshed both of its compact desktops, and they land on 22 September 2026 — pre-orders are open right now. The headline is the Mac Studio with the new M5 Max and M5 Ultra chips, and it's joined by a freshly Mac mini built around the new M6 and M5 Pro chips. Here's what actually changed and whether any of it matters for you.
The short version
Before the detail, here's the lineup at a glance.
Model Chips CPU cores GPU cores Memory ───────────────────────────────────────────────────────────────────────────────────────────── Mac mini (2026) M6 / M5 Pro 12 / up to 18 12 / up to 20 16–64GB Mac Studio (2026) M5 Max / M5 Ultra 18 / up to 36 up to 40 / 80 36–512GB
Both are available from 22 September, with the top-tier 512GB M5 Ultra option arriving in late October. That's the only genuinely "still coming" piece — everything else is on the shelf by late September.
Mac Studio: the real deal
The Mac Studio has always been the "small box, enormous performance" option — the one you buy when the Mac mini tops out but you don't want a tall Mac Pro. This year it gets two chips, and both are among the most powerful Apple silicon ever made.
M5 Max — the creator's sweet spot
- 18-core CPU, up to a 40-core GPU
- Up to 128GB unified memory, up to 614GB/s bandwidth
- 16-core Neural Engine
- Up to 2.5× faster CPU, 4.9× faster GPU, and 10.7× faster AI than the M1 Max
- Up to 5 displays, plus 19 simultaneous streams of 8K ProRes 422 video
M5 Max is aimed at editors, developers, and designers. The AI numbers are the standout — a 10.7× jump in AI performance over M1 Max means running large language models locally feels dramatically more responsive, and the third-gen ray-tracing engine makes 3D and gaming on Mac genuinely more competitive.
M5 Ultra — the ceiling
M5 Ultra fuses two M5 Max chips into a single system-on-chip. That's Apple's trick for the top end: combine two of the best and call it Ultra. The result:
- Up to a 36-core CPU (the most cores ever in a Mac) and up to an 80-core GPU
- Up to 512GB unified memory at a massive 1.2TB/s bandwidth
- 32-core Neural Engine
- Up to 2.4× faster CPU, 4.7× faster GPU, and 9.8× faster AI than the M1 Ultra
- Up to 8 displays, plus 33 simultaneous streams of 8K ProRes 422 video
This is the chip for colour-grading uncompressed 8K, running computational fluid dynamics, or training and running frontier-class models entirely on device. Note that the full 512GB configuration arrives in late October; the rest of the range ships on 22 September.
What's genuinely new under the hood
Beyond the core counts, three things define this generation:
- Memory bandwidth. M5 Max reaches 614GB/s and M5 Ultra a staggering 1.2TB/s. Bandwidth is often what limits AI and video work, so this is where the big real-world gains come from.
- Thunderbolt 5. Up to 120 Gbps — roughly 3× the bandwidth of Thunderbolt 4. It handles external PCIe expansion, genlocked multi-display sync, and daisy-chaining, and Apple now supports clustering multiple Studios for up to 3× faster distributed AI inference.
- Connectivity. Both machines add Wi‑Fi 7 and Bluetooth 6, and the Mac Studio keeps its full pro I/O: Thunderbolt 5, SDXC, 10Gb Ethernet, and even USB‑A.
Mac mini: M6 and M5 Pro
The Mac mini — the most affordable Mac desktop — gets two chips too. It's the machine most people actually buy, so these updates reach more users.
- M6 — the new mainstream chip: 12-core CPU, 12-core GPU, a dual 16-core Neural Engine, up to 32GB memory at 170GB/s, and up to 4.8× faster graphics than the M4.
- M5 Pro — the pro opt-in: up to an 18-core CPU and a 20-core GPU, up to 64GB memory at 307GB/s, and up to 4× faster than the M4 Pro.
The mini stays tiny (12.7 × 12.7 cm, a 5-inch square) but now puts ports on both the front and back, adds Thunderbolt 5 on the M5 Pro model, 2.5Gb Ethernet, and supports up to three 6K displays with M5 Pro. It's the best value in the lineup by a wide margin.
Running LLMs locally
This is where the Mac Studio really earns its name, and it's probably why you're here. Apple's unified-memory design is unusually well suited to running large language models on device, and the MLX ecosystem — plus tools like LM Studio, Ollama, and exo — makes it easy. Two numbers decide almost everything:
- How much memory you have decides how big a model you can load. In a local LLM the model's weights must live in memory all at once. Apple's unified memory means the whole model, plus its context (the KV cache), shares one pool — so the largest model you can run is roughly the total memory minus a few gigabytes for the OS.
- Memory bandwidth decides how fast it generates. For large models, word-by-word generation is bottlenecked on pushing weights out of memory, so tokens-per-second track roughly the usable bandwidth. That's exactly why the Ultra's 1.2TB/s matters so much over the Max's 614GB/s.
Apple doesn't publish exact token rates, so the figures below are rough estimates based on memory bandwidth and how much space a quantized model takes. Real numbers depend on quantization (4-bit vs 8-bit), context length, and the tool you use — but they give a useful sense of what each machine can do.
Model (4-bit quant) ~size M5 Pro (Mac mini) M5 Max M5 Ultra ────────────────────────────────────────────────────────────────────────────────────────────── 14B ~7GB very fast ~40–80 tok/s ~60–100 tok/s 32B ~16GB ~15–25 tok/s ~25–40 tok/s ~45–70 tok/s 70B ~37GB ~8–15 tok/s ~15–25 tok/s ~35–55 tok/s 120B+ ~63GB won't fit * limited ** ~15–25 tok/s
* The Mac mini with M5 Pro tops out at 64GB, so a 70B model is about its ceiling and larger ones won't load. The base M6 mini (32GB) caps you around 30B.
A 70B model at 4-bit takes ~37GB, so it needs at least the 64GB M5 Max config to run comfortably — the 36GB baseline is better suited to 32B-class models. The 120B+ class needs the M5 Ultra's big memory, and the 512GB option arriving in October opens the door to even larger models, heavy batching, and very long context windows. For most people, a M5 Max with 64GB or more is the sweet spot for local AI: it runs 70B-class models at usable speeds without chasing the Ultra's price.
Which desktop is yours?
Use case Best choice ──────────────────────────────────────────────────── Home office, coding, web Mac mini with M6 Serious hobbyists, light pro work Mac mini with M5 Pro Video editing, design, dev Mac Studio with M5 Max Frontier AI, 8K, multi-display Mac Studio with M5 Ultra
For most people the Mac mini with M6 is more than enough — it's remarkably fast for the price. Step up to M5 Pro if you want headroom for creative work. Only reach for the Mac Studio if you've outgrown the mini; and only chase M5 Ultra if you're running models or 8K pipelines that genuinely need 512GB of memory.
Bottom line
This is a quiet-but-massive refresh. The Mac Studio now offers M5 Max and M5 Ultra — the strongest desktop Macs Apple has made, with AI performance and memory bandwidth that leap years ahead of the M1-era chips. The Mac mini with M6 and M5 Pro keeps the entry price attractive while pulling most of those upgrades down to the everyday machine.
Both ship on 22 September. If you're after the full 512GB M5 Ultra, wait for late October. Otherwise, pre-order now and pick the chip that matches your workload — you rarely want to under- or over-buy here.