Sunk Cost sunkcost.ai Data checked 2026-09-03

Can a Mac mini M6, 32GB run local LLMs?

Yes — 19 of the 39 open models on this site fit in its 21 GB of usable memory, the strongest being Qwen3.8 27B. Whether that saves you money is a different question, and the answer is usually no.

Price$1,299
Memory32 GB, about 21 GB of it addressable by the GPU at 170 GB/s
Best model it runsQwen3.8 27B — Sonnet-class, 7 tok/s
Pay-back against the APIPays back in 17 years at 500k tokens a day

Run the numbers on this machine

What it runs

ModelSpeedClassGood atMemory
Qwen3.8 27BQ4_K_M 6.9 tok/s Sonnet-class 19 GB
Qwen3.6 27BQ4_K_M 6.7 tok/s Haiku-class 19 GB
Muse Glimmer 30BQ4_K_M 7.2 tok/s Haiku-class 18 GB
Gemma 4 26B-A4BQ4_K_M 15 tok/s Haiku-class 18 GB
GLM-4.7-FlashQ4_K_M 14 tok/s Haiku-class 20 GB
Gemma 4 12BQ4_K_M 16 tok/s Below every hosted tier 8.0 GB
Qwen3.5 9BQ4_K_M 19 tok/s Below every hosted tier 6.8 GB
MiniCPM5 2BQ4_K_M 43 tok/s Below every hosted tier 3.0 GB
Qwen3.5 4BQ4_K_M 33 tok/s Below every hosted tier 3.8 GB
Ling 3.0 tinyQ4_K_M 21 tok/s Below every hosted tier 6.4 GB
Granite 4.2 8BQ4_K_M 12 tok/s Below every hosted tier 11 GB
gpt-oss-20bMXFP4 18 tok/s Below every hosted tier 13 GB

7 more fit; the calculator lists them all.

The specifics

Chip
M6 — 12-core CPU / 12-core GPU
Memory bandwidth
170 GB/s
Usable by the GPU
21 GB — macOS lets the GPU wire roughly two-thirds of unified memory by default (llama.cpp discussion #2182: 66.7% at 32 GiB or less, 75% above). Raise it with `sudo sysctl iogpu.wired_limit_mb`. Memory is treated as GB throughout, which is slightly conservative.
Power under load
65 W (stand in) — Apple has not published power figures for the 2026 machines yet (support articles 103253 / 102027 stop at the 2024/2025 models). Stand-in: Apple's published maximum for the previous chip (M4 Mac mini, 65 W), which is a CPU+GPU stress figure, so it likely overstates inference draw.
Availability
Pre-order; ships 2026-09-22.
Sources
source 1, source 2, source 3