Benable — create shareable lists of things you recommend!

E.g., products you love, local businesses, travel recs - you can add anything to a Benable list!

Discover

Best Mini PCs and Macs for Local LLMs 2026

16 items

Lifemaxxing

Small, quiet boxes for running open models at home, grouped by what they can hold. Unified-memory machines let you load 70B and 120B mixture-of-experts models that no single consumer GPU can, but they run them slower than a big GPU. Notes cite The Ve...

Sections

8B to 32B models: compact and quiet

70B and 120B MoE: 128GB Strix Halo

CUDA boxes: DGX Spark class

Fastest dense models: Mac Studio

8B to 32B models: compact and quiet

Always-on boxes for everyday chat and coding models.

Fits 7B to 14B models with 307 GB/s of memory bandwidth, starting at $1,699 per The Verge. Pick 64GB for 30B-class models. Source: The Verge.

Apple Mac mini M5 Pro 24GB

Pick this if you want the cheapest Mac that holds 30B-class models comfortably. This config has 64GB of unified memory, an 18-core CPU, a 20-core GPU and a 1TB SSD.

Apple Mac mini M5 Pro 64GB

Pick this for a quiet, low-power box that runs small chat and coding models. It topped one llama.cpp efficiency test and still handled a 14B coding model.

Apple Mac mini M6 24GB

Pick this if you want a cheaper Strix Halo box for mid-size models. The GPU can use up to 48GB here, but expect slower prompt reading than NVIDIA's GB10 and some ROCm rough edges.

Corsair AI Workstation 300 64GB

70B and 120B MoE: 128GB Strix Halo

About 96GB usable as VRAM: room for 70B and gpt-oss 120B for the least money.

Cheapest common 128GB box for gpt-oss 120B, about 34 tokens/sec in community tests; dense 70B only manages 4.5 to 5, and prompts read slowly. Source: lhl Strix Halo tests; Compute Market.

GMKtec EVO-X2 128GB

The business-grade Strix Halo box with ECC and a 3-year warranty; StorageReview ran gpt-oss 120B on it with no discrete GPU. Source: DataHardware.

HP Z2 Mini G1a

Pick this if you want a 128GB Strix Halo box from a big brand that runs 30B coding models offline at a quick pace. The downsides are slow 2.5GbE networking and a high price.

MSI Pro Max Edge AI+ 128GB

Pick this if you want AMD's top Strix Halo chip with 128GB for well under the price of AMD's own Halo box. Up to 96GB can go to the GPU, though networking is only 2.5GbE.

Bosgame M5 128GB

CUDA boxes: DGX Spark class

128GB with NVIDIA's full CUDA stack, for fine-tuning and faster prompt reading.

Best for NVIDIA's CUDA tools in a small box: Dell's version of the DGX Spark, with the same GB10 chip and 128GB unified memory. LMSYS measured about 50 tokens/sec on gpt-oss 120B with SGLang on the DGX Spark. Dense 70B is slow because bandwidth is th...

Dell Pro Max with GB10

Pick this if you want ASUS's take on the DGX Spark. It uses the same GB10 chip with 128GB of memory and full CUDA support, and suits CUDA work and clustering.

ASUS Ascent GX10

Pick this if you live in NVIDIA's tools and want HP's version of the DGX Spark. Big models start fast but slow sharply as the context fills, and idle power is noticeable.

HP ZGX Nano G1n 128GB

Pick this if you want a GB10 Spark box that runs cooler than its rivals, with inference speed matching the other Spark systems. The linked model is the 2TB Linux version.

Acer Veriton GN100 128GB

Fastest dense models: Mac Studio

Apple bandwidth is what makes dense 70B usable on a desktop.

Starts at $2,499 per The Verge, and llamaperf says an M5 Max covers most local chat use if your models fit in 128GB. The linked 36GB base is for 14B to 27B; pick more memory for 70B. Source: The Verge; llamaperf.

Apple Mac Studio M5 Max

The one to get if dense 70B or 120B MoE is the goal: 18-core CPU, 40-core GPU, 128GB unified memory with 614GB/s bandwidth, 1TB SSD. macOS gives the GPU roughly three quarters of memory. Source: llamaperf; B&H specs.

Apple Mac Studio M5 Max 128GB

Pick this if your models fit in about 72GB and you want them to run faster than on a 128GB M5 Max. This is the 64-core GPU Ultra with 96GB, and it is a preorder for now.

Apple Mac Studio M5 Ultra 96GB

Pick this if you need more than 128GB on a Mac, since the Ultra is the only one that goes higher. About 192GB of this 256GB config goes to models by default, and it is a preorder for now.

Apple Mac Studio M5 Ultra 256GB