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

Can a Mac Studio M5 Ultra, 96GB run local LLMs?

Yes — 29 of the 39 open models on this site fit in its 72 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$5,499
Memory96 GB, about 72 GB of it addressable by the GPU at 1200 GB/s
Best model it runsQwen3.8 27B — Sonnet-class, 48 tok/s
Pay-back against the APIPays back in 68 years at 500k tokens a day

Run the numbers on this machine

What it runs

ModelSpeedClassGood atMemory
Qwen3.8 27BQ4_K_M 48 tok/s Sonnet-class 19 GB
Qwen3.6 27BQ4_K_M 47 tok/s Haiku-class 19 GB
Qwen3.6 35B-A3BQ4_K_M 143 tok/s Haiku-class 23 GB
Muse Glimmer 30BQ4_K_M 50 tok/s Haiku-class 18 GB
Gemma 4 26B-A4BQ4_K_M 105 tok/s Haiku-class 18 GB
Gemma 4 31B itQ4_K_M 35 tok/s Haiku-class 26 GB
Granite 4.2 30BQ4_K_M 34 tok/s Haiku-class 26 GB
GLM-4.7-FlashQ4_K_M 102 tok/s Haiku-class 20 GB
Gemma 4 12BQ4_K_M 113 tok/s Below every hosted tier 8.0 GB
Qwen3.5 9BQ4_K_M 133 tok/s Below every hosted tier 6.8 GB
Nemotron 3.5 Lightning 30B-A3BQ4_K_M 137 tok/s Below every hosted tier 26 GB
MiniCPM5 2BQ4_K_M 303 tok/s Below every hosted tier 3.0 GB

17 more fit; the calculator lists them all.

The specifics

Chip
M5 Ultra — 30-core CPU / 64-core GPU
Memory bandwidth
1200 GB/s
Usable by the GPU
72 GB — macOS lets the GPU wire roughly 75% 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
270 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 (M3 Ultra Mac Studio, 270 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