Apple · M2 Ultra · Desktop

Best AI Models for Mac Studio M2 Ultra (192 GB)

Runs 27–31B and 120B MoE models well; dense 70B fits but is slow at 12 tok/s.

192 GB unified188.5 GB for models800 GB/s54 TFLOPS FP16295 W2023

Buy$5,599

Memory

64 GB128 GB192 GB$5,599

Same chip in

Mac Studio$5,599 Mac Pro

What LLMs can the Mac Studio M2 Ultra run?

#	Model	Speed	Quant · Memory
1	Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 32.1 t/s	32.1 tok/s	Q4_K_M · 17.9 GB
2	Inkling SmallThinking Machines Q4_K_M · 162.7 GB · 97.6 t/s	97.6 tok/s	Q4_K_M · 162.7 GB
3	Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 32.1 t/s	32.1 tok/s	Q4_K_M · 17.9 GB
4	MiniMax M2.7MiniMax Q4_K_M · 139.6 GB · 41.2 t/s	41.2 tok/s	Q4_K_M · 139.6 GB
5	NVIDIA Nemotron 3.5 Lightning 30B A3B BF16NVIDIA Q4_K_M · 19.3 GB · 77.0 t/s	77.0 tok/s	Q4_K_M · 19.3 GB

#	Model	Speed	Quant · Memory
1	Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 32.1 t/s	32.1 tok/s	Q4_K_M · 17.9 GB
2	Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 27.5 t/s	27.5 tok/s	Q4_K_M · 21.7 GB
3	Qwen3.6 35B A3BAlibaba Q4_K_M · 22.0 GB · 75.9 t/s	75.9 tok/s	Q4_K_M · 22.0 GB
4	Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 91.7 t/s	91.7 tok/s	Q4_K_M · 16.5 GB
5	NVIDIA Nemotron 3.5 Lightning 30B A3B BF16NVIDIA Q4_K_M · 19.3 GB · 77.0 t/s	77.0 tok/s	Q4_K_M · 19.3 GB

#	Model	Speed	Quant · Memory
1	Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 32.1 t/s	32.1 tok/s	Q4_K_M · 17.9 GB
2	Inkling SmallThinking Machines Q4_K_M · 162.7 GB · 97.6 t/s	97.6 tok/s	Q4_K_M · 162.7 GB
3	Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 27.5 t/s	27.5 tok/s	Q4_K_M · 21.7 GB
4	Qwen3.5 27BAlibaba Q4_K_M · 17.9 GB · 32.1 t/s	32.1 tok/s	Q4_K_M · 17.9 GB
5	NVIDIA Nemotron 3.5 Lightning 30B A3B BF16NVIDIA Q4_K_M · 19.3 GB · 77.0 t/s	77.0 tok/s	Q4_K_M · 19.3 GB

#	Model	Speed	Quant · Memory
1	Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 32.1 t/s	32.1 tok/s	Q4_K_M · 17.9 GB
2	Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 32.1 t/s	32.1 tok/s	Q4_K_M · 17.9 GB
3	MiniMax M2.7MiniMax Q4_K_M · 139.6 GB · 41.2 t/s	41.2 tok/s	Q4_K_M · 139.6 GB
4	Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 27.5 t/s	27.5 tok/s	Q4_K_M · 21.7 GB
5	NVIDIA Nemotron 3.5 Lightning 30B A3B BF16NVIDIA Q4_K_M · 19.3 GB · 77.0 t/s	77.0 tok/s	Q4_K_M · 19.3 GB

#	Model	Speed ↓	Quant · Memory
1	Qwen3.5 4BAlibabaHow to run → Q4_K_M · 3.4 GB · 172 t/s	172 tok/s	Q4_K_M · 3.4 GB
2	GPT OSS 20BOpenAI Q4_K_M · 13.1 GB · 106 t/s	106 tok/s	Q4_K_M · 13.1 GB
3	Inkling SmallThinking Machines Q4_K_M · 162.7 GB · 97.6 t/s	97.6 tok/s	Q4_K_M · 162.7 GB
4	Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 91.7 t/s	91.7 tok/s	Q4_K_M · 16.5 GB
5	NVIDIA Nemotron 3.5 Lightning 30B A3B BF16NVIDIA Q4_K_M · 19.3 GB · 77.0 t/s	77.0 tok/s	Q4_K_M · 19.3 GB

#	Model	Speed	Quant · Memory
1	Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 32.1 t/s	32.1 tok/s	Q4_K_M · 17.9 GB
2	MiniMax M2.7MiniMax Q4_K_M · 139.6 GB · 41.2 t/s	41.2 tok/s	Q4_K_M · 139.6 GB
3	MiniMax M2MiniMax Q4_K_M · 139.6 GB · 41.2 t/s	41.2 tok/s	Q4_K_M · 139.6 GB
4	Qwen2.5 Coder 32B InstructAlibaba Q4_K_M · 24.3 GB · 27.3 t/s	27.3 tok/s	Q4_K_M · 24.3 GB
5	NVIDIA Nemotron 3.5 Lightning 30B A3B BF16NVIDIA Q4_K_M · 19.3 GB · 77.0 t/s	77.0 tok/s	Q4_K_M · 19.3 GB

See all 411 models on the Mac Studio M2 Ultra →Bars = traction for models without a Score yet

About the Mac Studio M2 Ultra (192 GB)

The Mac Studio M2 Ultra runs 27–31B and 120B MoE models well. macOS keeps about 3.5 GB, leaving 188.5 GB for models. Prompt reading is its weak spot: a 4K-token prompt waits 21 s before the first word on Qwen3.8 27B. At its $5,599 MSRP it gives 13.6 tok/s per $1,000, 3rd of the 4 machines compared here. M2 Ultra generation, released 2023; the newer M3 Ultra writes 1% faster with 1.2× the compute.

For models188.5 GBmacOS keeps 3.5 GB

ChipM2 Ultrareleased 2023

Mac Studio M2 Ultra strengths and weaknesses

Strengths

Runs 120B MoE models: GPT OSS 120B at 74.6 tok/s in 70.8 GB
Room for long context: Qwen3.8 27B at Q4 uses 17.9 GB, leaving 170.6 GB free

Weaknesses

Slow prompt reading: A 32K-token prompt on Qwen3.8 27B waits 2.8 min for the first token
No CUDA: vLLM, AWQ, GPTQ and FP8 builds don't run; use GGUF or MLX

Memory for models

188.5 GB of 192 GB

35B MoE22 GB70B51.3 GB120B MoE70.8 GB309B186.5 GB

120B MoE70.8 GB309B186.5 GB

For models 188.5 GBmacOS keeps 3.5 GBModel size at Q4

Everything on the size ladder fits. Two 27B-class models together use 35.7 GB and leave 152.8 GB free.

Speed per dollar and per watt

NewUsed

On Qwen3.6 35B A3B (35B params, 3B active) at Q4_K_M. New means a shop price, either a street quote or the maker's own price; used is compared with used. The last two columns are the speed you get per $1,000 of price and per 100 W of power; higher is better. The Mac Studio M2 Ultra has no current new price, so it is shown at its MSRP.

Machine	New price	tok/s	tok/s per $1,000	tok/s per 100 W
RTX 509032 GB	about $5,376	213	39.7	37.1
Mac Studio M4 Max128 GB	about $3,499	64.2	18.3	44.3
Mac Studio M2 Ultra192 GB	about $5,599	75.9	13.6	25.7
Mac Studio M3 Ultra256 GB	about $7,099	76.6	10.8	28.4

Machine	Used price	tok/s	tok/s per $1,000	tok/s per 100 W
RTX 409024 GB	about $3,000	151	50.3	33.6
RTX 509032 GB	about $6,400	213	33.3	37.1

Cost per token: the Mac Studio M2 Ultra against 24–256 GB machines and GPUs

On Qwen3.6 35B A3B (35B params, 3B active, Q4_K_M, 4K context). Speed, cost per token and power per token depend on the model.

24–36 GB96–128 GB192+ GBthis desktopat MSRP

Cost per 1M output tokenson Qwen3.6 35B A3B · card over 3 years + power

$4.84

Machine, MSRP: $4.68
Power: $0.16

3-year cost of ownership$5,793

Or rent an RTX PRO 6000 Blackwell Max-Q Workstation Edition in the cloud$1.07/hr

Renting costs as much as buying after ≈5,200 hours (≈3.6 years at 4 h a day).

Include card price

Power$/kWhUseh/dayKeepyears

Each dot is a machine or GPU; hover or tap it for its numbers, click to open it. Machines without a price for the chosen basis are left out. How we calculate cost per token

Similar devices and alternatives

Other memory sizes of the same machine first, then machines and GPUs people weigh against it, each with the one reason it is worth a look. Speeds on Qwen3.6 35B A3B.

Same chip, 64 GB less memoryMac Studio M2 Ultra (128 GB)Apple · 124.5 GB for models · M2 Ultra

same speed▼ 64 GBno price

Same family Same chip, 128 GB less memoryMac Studio M2 Ultra (64 GB)Apple · 60.5 GB for models · M2 Ultra

same speed▼ 128 GBno price

Same family Newer machine, as fastMac Studio M3 Ultra (256 GB)Apple · 252.5 GB for models · M3 Ultra

same speed▲ 64 GB$7,099

Newer machine, 15% slowerMac Studio M4 Max (128 GB)Apple · 124.5 GB for models · M4 Max

▼ 15% speed▼ 64 GB$3,499

2.0× faster if your models fit in 24 GBRTX 4090NVIDIA · 24 GB VRAM · GPU

▲ 99% speed▼ 164.5 GB$1,599rent $0.34/hr

2.8× faster if your models fit in 32 GBRTX 5090NVIDIA · 32 GB VRAM · GPU

▲ 181% speed▼ 156.5 GB$5,376rent $0.50/hr

Mac Studio M2 Ultra tokens per second by model size

Output speed depends on memory bandwidth (800 GB/s). Grey rows need more than the 188.5 GB available.

9BQwen3.5 9B · Q4_K_M · 6.6 GB
88.1tok/s
14BQwen3 14B · Q4_K_M · 11.9 GB
58.9tok/s
27–31BQwen3.8 27B · Q4_K_M · 17.9 GB
32.1tok/s
35B MoEQwen3.6 35B A3B · Q4_K_M · 22.0 GB
75.9tok/s
70BLlama 3.3 70B Instruct · Q4_K_M · 51.3 GB
12.0tok/s
120B MoEGPT OSS 120B · Q4_K_M · 70.8 GB
74.6tok/s

Estimated speeds: Q4_K_M, 4K context, one user. How we estimate

Ultra generations: writing vs compute

M1 Ultra → M3 Ultra

Writing speed (tok/s)Compute (TFLOPS FP16)

Writing speed follows memory bandwidth and barely moved: 75.9 to 76.6 tok/s since the M1 Ultra. Compute, which sets how fast prompts are read, grew 2.1×. Indexed to the M1 Ultra; speeds on Qwen3.6 35B A3B, same 128 GB.

All 1309 models on the Mac Studio M2 Ultra

Featured only

Best quant that fits in 188.5 GB, ordered by llmrun Score. Fit grades how well it runs here, S best. How ranking works

LLM models compatible with Mac Studio M2 Ultra (192 GB) — ranked by Score, recency and Fit
Model	Quant	VRAM	Speed	Context	Fit
Dolphin Mistral 24B Venice Edition24.0BChat Q4_K_M·37.0 t/s	Q4_K_M	18.1 GB10%	37.0 t/s	131K	A74, Runs well
GPT OSS BF167MChatFunctions Q4_K_M·1866.7 t/s	Q4_K_M	0.3 GB0%	1866.7 t/s	131K	S100, Runs great
Qwen3.5 9B Uncensored9BChat Q4_K_M·94.3 t/s	Q4_K_M	8.6 GB5%	94.3 t/s	—	S90, Runs great
Qwen3 235B A22B235.1BChatFunctions Q4_K_M·23.4 t/s	Q4_K_M	142.9 GB76%	23.4 t/s	41K	B64, Decent
Kappa 20B 131k20.9BChat Q4_K_M·106.1 t/s	Q4_K_M	13.1 GB7%	106.1 t/s	131K	S92, Runs great
Qwen3.6 28B28.2BChat Q4_K_M·84.9 t/s	Q4_K_M	17.4 GB9%	84.9 t/s	262K	S88, Runs great
LightOnOCR 3 0.8B853MVisionFunctions Q4_K_M·651.2 t/s	Q4_K_M	0.9 GB0%	651.2 t/s	262K	S100, Runs great
MIST Mini 8B Thinking8.0BChatReasoning Q4_K_M·103.9 t/s	Q4_K_M	7.3 GB4%	103.9 t/s	131K	S92, Runs great
Qwen3 8B8BChat Q4_K_M·106.1 t/s	Q4_K_M	7.8 GB4%	106.1 t/s	—	S92, Runs great
Moondream3 Preview9.3BVision BF16·27.5 t/s	BF16	23.1 GB12%	27.5 t/s	—	A68, Runs well
Qwen3 4B Instruct 2507 Heretic4.0BChat Q4_K_M·193.1 t/s	Q4_K_M	4.2 GB2%	193.1 t/s	262K	S99, Runs great
Tiny Random LlamaForCausalLM1MChat Q4_K_M·1866.7 t/s	Q4_K_M	0.3 GB0%	1866.7 t/s	8K	S100, Runs great
JOSIE 1.1 4B Instruct4.0BChat Q4_K_M·193.1 t/s	Q4_K_M	4.2 GB2%	193.1 t/s	262K	S99, Runs great
CAT Thinking 8B8.2BChat Q4_K_M·101.4 t/s	Q4_K_M	7.6 GB4%	101.4 t/s	41K	S92, Runs great
Recursive Language Model 198M198MChat BF16·1272.7 t/s	BF16	0.4 GB0%	1272.7 t/s	1K	S100, Runs great

Mac Studio M2 Ultra specs and power

Specs

Chip: M2 Ultra
CPU cores: 24
GPU cores: 76
Unified memory: 192 GB
For models: 188.5 GB (macOS keeps 3.5 GB)
Memory bandwidth: 800 GB/s
FP16 (dense): 54.4 TFLOPS
Neural Engine: 32 TOPS
Power: 295 W max
Released: 2023-06-13
MSRP: $5,599

Mac Studio M2 Ultra FAQ

Is the Mac Studio M2 Ultra (192 GB) worth it for local LLMs in 2026?

At its $5,599 MSRP it gives 13.6 tok/s per $1,000 on Qwen3.6 35B A3B, 3rd of the 4 machines compared here. It runs 27–31B and 120B MoE models well; dense 70B fits but is slow at 12 tok/s.

How much memory is available for AI on the Mac Studio M2 Ultra (192 GB)?

About 188.5 GB of the 192 GB. macOS keeps roughly 3.5 GB for itself and other apps.

What is the largest model the Mac Studio M2 Ultra (192 GB) can run?

MiMo V2.6 Flash RL (309B) is the largest at 4-bit or better: at Q4_K_M it uses 186.5 GB, 99% of the available memory, and runs at about 38.9 tok/s.

Can the Mac Studio M2 Ultra (192 GB) run Qwen3.8 27B?

Yes. Qwen3.8 27B fits at Q4_K_M, using 17.9 GB of the 188.5 GB available, and it is the best model that runs well on this machine.

How long does the Mac Studio M2 Ultra (192 GB) take to read a long prompt?

On Qwen3.8 27B, about 21 s for a 4K-token prompt and 2.8 min for a 32K-token document before the first word appears.

Is renting a GPU cheaper than buying the Mac Studio M2 Ultra?

An RTX PRO 6000 Blackwell Max-Q Workstation Edition (96 GB) rents from $1.07/hr. Renting costs as much as buying after ≈5,200 hours (≈3.6 years at 4 h a day).

48 GB GPUs All Macs Desktops & mini PCs Best coding LLMs All hardware