Apple Silicon for Local LLMs — what to buy in October 2026

Ranked buy-now

1. 1. Mac Studio M5 Max, 36GB / 512GB

   $2,449.00

   In stock, Amazon.com · ASIN B0HGKSQMX6

   Best Amazon-sold LLM workstation in this check. Current-gen Studio. The Amazon title and specs say 36GB unified memory (an 18-core CPU / 32-core GPU in the title). A deal note elsewhere said 32GB; use the Amazon listing.

   Model fit. Q4 ~16–32B dense. Qwen3.8-27B and K2-Horizon-32B / 36B-A4B, with KV headroom. Not a 70B+ MoE box.

   Check B0HGKSQMX6 on Amazon

2. 2. Mac mini M5 Pro, 24GB / 512GB

   $1,669.99

   In stock, Amazon.com · ASIN B0HGGHNQY6

   Best compact Pro mini. Listing copy cites 307 GB/s memory bandwidth and a 15-core CPU / 16-core GPU.

   Model fit. Q4 ~7–27B. K2-Horizon-7B is easy. Qwen3.8-27B is tight with the OS and KV cache. Skip 32B+ dense.

   Check B0HGGHNQY6 on Amazon

3. 3. Mac mini M4 Pro, 24GB / 512GB

   $1,569.99

   Sold by Amazon.com · ASIN B0DLBVHSLD

   Prior-gen Pro mini still stocked. 12-core CPU / 16-core GPU and 273 GB/s class bandwidth. Slightly cheaper than the M5 Pro 24GB on Amazon that day.

   Model fit. Same 24GB Q4 band as the M5 Pro mini: ~7–27B, with 27B tight.

   Check B0DLBVHSLD on Amazon

4. 4. GMKtec EVO-X2, 128GB (Strix Halo)

   $3,649.99

   In stock, sold by GMKtec-US · ASIN B0F53MLYQ6

   Default Strix Halo buy, not the Beelink GTR9 Pro. Same Max+ 395 and ~256 GB/s, about $700 less, single 2.5GbE. Up to 96 GB of the 128 GB pool can be assigned as VRAM. Not CUDA.

   Model fit. 96–128 GB at ~256 GB/s: DeepSeek-V4-Flash aggressive quant, Qwen3.8-Flash-Next MoE, dense 70B Q4 slow.

   Check B0F53MLYQ6 on Amazon

Catalog entries live on the hardware index. Compare any two of them on the compare page. Check a specific model’s footprint with the VRAM calculator.

Memory tiers and the models that match them

Model names below are from the open-weight window 29 May 2026 through 1 October 2026 (Qwen3.8, K2-Horizon, DeepSeek-V4-Flash, GLM-5.3-Flash, MiniMax-M3, Qwen3.8-Flash-Next). They are mapped to hardware by the Q4 rule, not by a fresh llama-bench run.

Not on Amazon — configure at Apple

If the model you care about is a large MoE, the in-stock Studio is the wrong box. These configs were on Apple’s buy page and did not have a US Amazon ASIN I could verify. An EU tracker ASIN (B0HGRN2PV2) returned 404 on amazon.com, so it is not listed.

• Mac Studio M5 Max 64GB / 1TB (an 18-core CPU / 40-core GPU variant appears on the Apple buy page). Mid-high unified memory for 32B+ and smaller MoEs.
• Mac Studio M5 Ultra 96GB / 1TB (30-core CPU / 64-core GPU). The large-MoE / DeepSeek-class local target. Direct configure link: M5 Ultra 96GB at Apple.
• Higher Ultra (256GB and above). Frontier open MoE such as GLM-5.3-Flash and MiniMax-M3. Order from the same Apple Studio page. No Amazon ASIN.

Start at apple.com/shop/buy-mac/mac-studio. The 36GB M5 Max base is also sold on Amazon as B0HGKSQMX6; the higher memory steps are the Apple-only part.

Watchlist, not buy-now

• Mac mini M4 Pro 48GB. B0DS2XP86K (14-core CPU / 20-core GPU, 1TB) and B0DYFCQ7CB (12-core CPU / 16-core GPU, 1TB) were both currently unavailable. No price is published. Useful if they return; not a recommendation today.
• Mac Studio M4 Max 64GB / 1TB. B0FNS1ZX5B had no featured offer. The prior-gen Studio base at 36GB (B0F5SNBZVF) was also unavailable and is superseded by the in-stock M5 Max 36GB when you are buying now.
• Mac Studio M4 Max 128GB. The catalog still has ASIN B0FMFZ5SXS. That price was not re-checked on 2026-10-01, so it is blank. Prefer the Apple configure page for 64GB and above.
• Base M4 16GB and 24GB. ASINs B0DLBTPDCS and B0DLBV145M are real and stay in the catalog, with price set to null. Both pages showed a “High price” badge and no clean buy box. A 16GB mini is a ~3–7B machine after the OS takes its share. Third-party B0DLBX4B1K and Renewed B0DTPPBN95 are marketplace offers, not primary buys.

Strix Halo when the Mac memory ceiling is the problem

Community discussion in September and October 2026 kept pointing at high unified memory — Studio M5 Max and Ultra, Strix Halo 128GB, Framework Desktop 192GB — for Qwen3.8, Flash-Next, and Muse. The default Amazon PC in that class is the GMKtec EVO-X2: 128 GB at about 256 GB/s, $3,649.99. The Beelink GTR9 Pro is the same APU and bandwidth with a higher price and a 10GbE NIC history; it is a caveat on the Strix Halo guide, not a second default. Plan on Linux or a ROCm/Vulkan stack. CUDA workflows still belong on an NVIDIA card; see the GPU buying guide and the Mac vs PC guide.

DeepSeek’s current big local target in this window is DeepSeek-V4-Flash (~155GB at a normal 4-bit). That is an EVO-X2 aggressive quant or an Apple Ultra, not a 24GB mini. Qwen sizing for the dense 27B class is covered in the Qwen hardware guide.

How to run whatever you buy

On the Macs, Ollama is the always-on worker (brew install ollama, then a model pull). LM Studio is the GUI. llama.cpp with Metal is the same engine with the flags exposed. MLX (pip install mlx-lm) is Apple’s framework and the one to reach for when you want to fine-tune on the GPU you already bought. Setup steps are in How to run LLMs on Mac and the Ollama cheat sheet. The quantization trade itself is in the quantization guide and how much VRAM you need.

A Mac mini in this tier is a reasonable always-on machine for agents and a self-hosted runner. It is not a substitute for 64GB+ when the weights do not fit. Memory is soldered. Buy the pool you need on day one.

Frequently asked questions