Mac mini vs Mac Studio for local AI
Buy for unified memory first and bandwidth second. As of 19 October 2026 a Mac mini tops out at 64GB and 307GB/s (M5 Pro, from $1,699); a Mac Studio starts at 36GB and 460GB/s for $2,499 on M5 Max, reaches 128GB on the 614GB/s chip that starts at $3,099, and the M5 Ultra starts at 96GB and 1.2TB/s for $5,499 and configures to 256GB, or to 512GB from late October, Apple's buy page says. Memory can't be upgraded later, so the largest model the machine will ever run is decided at checkout.
Check the date. The specs and prices on this page were read from Apple on 19 October 2026. Apple's buy pages no longer sell the M4 mini, M4 Max Studio or M3 Ultra Studio, so this page covers the machines Apple sells now; the older chips are covered on our M1 vs M4 page. What does not change: memory cannot be upgraded later, so buy for the largest model you want to run, and re-check Apple's pages before you order.
Outlier Pro: $49 once · Black Friday price, all of November: $24.50 once, until 30 November
Free to start (Nano and Lite). macOS 15+, Apple silicon. Refund window: 30 days, no questions asked.
I build Outlier, a local-AI app for Apple Silicon, so I have an obvious interest in you buying one of these. I'll be useful anyway, including where a PC with a discrete GPU is the better buy. Every spec and price came off Apple's own pages on 19 October 2026.
Capacity and bandwidth are two different problems
Running a model locally makes two demands. Capacity: weights, KV cache and OS have to fit in memory, or you wait on the SSD. Bandwidth: generation reads a large fraction of those weights for every token, so decoding is memory-bound and bandwidth predicts tokens per second far better than core count.
Apple silicon is good at the first: CPU and GPU share one pool, with no separate VRAM budget and no host-to-device copy, which is the assumption Apple's MLX framework is built on. Whatever RAM you bought is GPU memory. On the second, the spread is enormous: 153GB/s on the base M6 mini, 307GB/s on M5 Pro, 460GB/s on the 32-core-GPU M5 Max (614GB/s with the 40-core GPU), 1.2TB/s on M5 Ultra. About eight-to-one, across two lines that look alike on a spec sheet.
What each configuration actually fits
Gigabytes are hard to reason about abstractly, so here are measured numbers from Outlier's tiers, the example I know best. Disclosure on the yardstick: Nano (4B) and Lite (9B) are free, while Quick, Core, Vision 3.8 and Plus require Outlier Pro, so most of the tiers below need Pro, which is a one-time purchase. RAM figures are minimums, with headroom for context and the OS.
- 16GB (base M6 mini): a 4B model is a 2.37GB download needing 6GB of RAM; a 9B model is 5.04GB needing 12GB. A 26B MoE model at 15.61GB has a 16GB minimum: it runs, but that's the floor.
- 24GB (M6 mini upgraded, M5 Pro mini base): dense 27B models become practical (15.13GB on disk, about 20.7 tok/s on an M1 Ultra) and the 27B dense Vision 3.8 model at 15.5GB fits.
- 64GB (M5 Pro mini at its top configuration; M5 Max Studio, 40-core GPU only): the floor for very large paged-MoE models. Outlier's 397B-a17b tier is a 209GB download streaming experts from SSD at 1.59 tok/s on my M1 Ultra with 64GB: batch speed, not conversational. Outlier lists 64GB as the minimum and 128GB as recommended for that tier; the Mac mini stops at that 64GB minimum, while the Mac Studio reaches 128GB on the 40-core-GPU M5 Max and more on the M5 Ultra.
- 96GB (M5 Ultra Studio base): the same models with headroom, at 1.2TB/s; the M5 Ultra configures to 256GB, and to 512GB from late October, Apple says.
Bandwidth decides how pleasant that is. A 4B model measures 71.7 tok/s on my M1 Ultra and 32 tok/s on an M4 MacBook Air (single prompt, 2026-05-04, indicative). A 24GB M6 mini should run a 27B model, but slower than the 20.7 tok/s I measured on an M1 Ultra; I have not measured it.
Side by side
| mini M6 | mini M5 Pro | Studio M5 Max | Studio M5 Ultra | |
|---|---|---|---|---|
| Starting price | $899 | $1,699 | $2,499 | $5,499 |
| Unified memory | 16GB → 32GB | 24GB → 64GB (18/20 chip, $1,899) | 36GB → 128GB (40-core GPU chip, $3,099) | 96GB → 256GB, 512GB from late October (80-core GPU chip, $6,799) |
| Memory bandwidth | 153 / 170GB/s | 307GB/s | 460 / 614GB/s | 1.2TB/s |
| Base storage | 256GB | 512GB | 512GB | 1TB |
| Max continuous power | 155W | 155W | 480W | 480W |
| Thunderbolt | 3× TB4 | 3× TB5 | 4× TB5 | 6× TB5 (4 back, 2 front) |
| $ per GB, base config | ~$56 | ~$71 | ~$69 | ~$57 |
| Largest Outlier tier (all Pro-only) | Quick 26B-a4b | Vision 3.8 27B | Plus 397B-a17b at 64GB | Plus 397B-a17b |
| Upgradeable later | No | No | No | No |
Thermals, noise, and sustained load
Inference is a sustained load, not a burst. Chat produces spikes; an agent loop or an overnight index pins the GPU for hours, the pattern a thermal envelope governs.
Apple rates maximum continuous power at 155W for the mini and 480W for the Studio, and lists the mini at 5 dBA idle. Loaded, you hear it. The mini will run models all day, but it's a small box with a small fan: I'd steer sustained work to a Studio even when the mini's memory fits.
Price per usable gigabyte
Starting price divided by base memory isn't linear: roughly $56/GB for the $899 M6 mini, $71 for the $1,699 M5 Pro mini, $69 for the $2,499 M5 Max Studio, and about $57/GB for the $5,499 M5 Ultra, which also has the bandwidth to use them. The middle is the expensive part.
Upgrade paths: there aren't any
Unified memory is configure-to-order with no user-serviceable path, so the ceiling is fixed at checkout and "I'll add RAM later" isn't a plan. What you can add is storage over Thunderbolt 5 at up to 120Gb/s: six ports on the M5 Ultra Studio, four on the M5 Max Studio, three on the M5 Pro mini, Thunderbolt 4 on the M6 mini. No substitute for memory. Networking is the same story as memory: 10Gb Ethernet is standard on the Studio and a configure-to-order option on the mini, so it has to be chosen at checkout too.
M5 Pro (up to 307GB/s, 64GB) and M5 Max (up to 614GB/s, 128GB) shipped in the MacBook Pro in March 2026, claiming over 4× the peak GPU compute for AI versus the previous generation. Both are in Apple's desktops now: M5 Pro in the Mac mini, M5 Max in the Mac Studio.
Where these machines genuinely win: and where they don't
The bad news first, because it's the part vendors skip. On raw throughput, a PC with a discrete GPU is usually faster. The fastest dedicated graphics memory runs at bandwidths these desktops don't reach (NVIDIA lists 1,792GB/s for the RTX PRO 6000), and the CUDA stack is deeper for training and batch serving. For any model that fits inside a card's VRAM, expect more tokens per second there.
The Apple case is capacity per dollar and power draw:
- Unified memory removes the VRAM cliff. NVIDIA lists the RTX 5090 at 32GB, so getting 96GB of GPU-addressable memory out of consumer cards means stacking them, with the power supply, chassis and noise that implies; NVIDIA's single-card route to 96GB is the 600W RTX PRO 6000.
- 1.2TB/s inside a 480W envelope for the whole machine; NVIDIA rates the RTX 5090 card alone at 575W.
- Everything is capital cost: no per-token billing, no rate limits, no egress, and it works with the network off, on a machine that still has AppleCare and resale value.
- The $899 M6 mini is the entry point: 16GB standard, a first-party ML stack in MLX, Metal and Core ML, and good third-party support in llama.cpp, Ollama and LM Studio.
One caveat for both: the hardware supplies no model, no license and no privacy guarantee, the software you install decides that. Apple Intelligence is itself hybrid, mixing on-device processing with Private Cloud Compute, so "it's a Mac" isn't an on-device claim.
Who should pick which
The $899 M6 Mac mini if you're finding out whether local AI fits your work. 4B and 9B models run briskly, a 26B MoE model sits at the memory floor, and if local isn't for you, you still own a good small computer.
The M5 Pro Mac mini (24–64GB) if you want a 27B-class daily driver, quietly and cheaply. 307GB/s is a real step up from 153, at $1,699 against the Studio's $2,499 entry price.
The M5 Max Mac Studio if the machine will be under sustained load (agent loops, batch jobs, video work alongside model work) or if you need 64GB or more, which requires the 40-core-GPU chip from $3,099.
The M5 Ultra Mac Studio only if 96GB or more at 1.2TB/s is the actual requirement. If your models fit in 64GB, this isn't your machine.
A PC with a discrete GPU if your model fits in VRAM, throughput is the goal, and you'll take on the power, noise and assembly.
Want to see what your Mac can actually run?
Outlier runs models fully on-device: no account, no caps, nothing uploaded. The free Nano and Lite tiers need 6GB and 12GB of RAM, so you can test before buying hardware.
Free: Nano + Lite. Pro: $49 once; Black Friday price, all of November: $24.50 once, until 30 November. macOS 15+. Refund window: 30 days, no questions asked.
Download free (276 MB) Buy ProOn your phone?