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

Yes — 34 of the 48 current models on this site fit in its 72 GB of usable memory. The strongest it holds, Qwen3.8 27B, runs at 48 tok/s with a 32k window, worked out from memory bandwidth. Whether that saves you money is a different question: at $5,499 and 500k tokens a day, it pays back in 307 years.

Run the numbers on this machine

What the Mac Studio M5 Ultra, 96GB runs

Memory never runs out first here. Every model below holds 256k, where the calculator's list ends, or the longest setting its own context limit allows, so the window you get is the model's choice rather than this machine's.

Speed and memory are at 32k context, the setting the calculator starts on; the memory column is the weights plus the cache for that much of it. Each speed says how it was arrived at, and every one here is estimated from this machine's 1200 GB/s of memory bandwidth rather than taken from a published benchmark. There is a page on how that sum works, and what each size needs, and one on what 96 GB runs, machine by machine. The longest context is the longest setting the calculator offers that this machine still holds the model at, cache included, and each one opens the calculator on that model at that length. A figure tagged memory is one this machine ran out of room for, and the rest are stopped by the model's own limit or by the end of the list.

For scale, the calculator starts from 80 tok/s for a hosted API and times a local machine against it. 7 of the 12 above reach it, and the slowest is 34 tok/s.

22 more fit. Ranked below the twelve above: MiniCPM5 2B, Qwen3.5 4B, gpt-oss-120b, Ling 3.0 tiny, Granite 4.2 8B, Qwen3-Coder 30B-A3B, Qwen3-Coder Next, gpt-oss-20b, Gemma 4 E4B, LFM2.5 2.6B, Devstral Small 2 24B, Ministral 3 14B, Ministral 3 8B. The other 9 have no intelligence-index score, so the ranking cannot place them: Laguna XS 2.1, LFM3 1.7B, Ornith 1.5 9B, Ornith 1.5 35B-A3B, Spark-X2.5 4B, Spark-X2.6 4B, KAT-Coder V2.5 Dev 35B-A3B, LFM2.5 24B-A2B, Flint 40B-A2.5B. Their pages show what each one needs and what runs it.

The 48 counted above are the current models. This site prices 16 superseded ones as well, because people still run them, and this machine holds 14 of those at 32k, strongest first: Qwen3 30B-A3B Instruct 2507, DeepSeek-R1-Distill-Qwen-32B, DeepSeek-R1-Distill-Llama-70B, Llama 3.3 70B Instruct, Llama 3.1 8B Instruct at Q4_K_M, Llama 3.1 8B Instruct at Q8_0, Mistral Small 3.2 24B Instruct, Qwen3 32B at Q4_K_M, Qwen3 32B at Q8_0, Qwen3 14B, Llama 4 Scout 17B-16E, Qwen3 8B, Gemma 3 27B it, Gemma 3 12B it. Each of those pages asks whether the model is still worth running.

Other machines to weigh against the Mac Studio M5 Ultra, 96GB

Every row uses the same defaults as the figures above: 500k tokens a day at 15:1 input to output, 32k context, and each machine's strongest model that fits, counted against the same 48 current models. The GeForce RTX 5090, 32GB is priced as the card alone, so add the PC around it before comparing it with a complete computer. All seven cards here are ranked by what each one holds.

Head to head with another computer: HP Z2 Mini G1a, 128GB · MacBook Pro M5 Max (16-inch), 64GB. With the same box and the bigger chip: 256GB. With the machine it replaced: Mac Studio M3 Ultra, 96GB. There are 111 machine match-ups on the site.

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

Apple (specs), Apple (newsroom), Daring Fireball