Apple · M5 Ultra · Desktop

Best AI Models for Mac Studio M5 Ultra (96 GB)

Runs 70–120B models well.

96 GB unified92.5 GB for models1.2 TB/s109 TFLOPS FP162026

Memory

96 GB$5,499 256 GB$9,499 512 GB$22,299

What LLMs can the Mac Studio M5 Ultra run?

#	Model	Speed	Quant · Memory
1	Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 49.3 t/s	49.3 tok/s	Q4_K_M · 17.9 GB
2	Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 49.3 t/s	49.3 tok/s	Q4_K_M · 17.9 GB
3	Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 42.2 t/s	42.2 tok/s	Q4_K_M · 21.7 GB
4	Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 109 t/s	109 tok/s	Q4_K_M · 16.5 GB
5	Muse Glimmer 30BMeta Q4_K_M · 19.6 GB · 46.9 t/s	46.9 tok/s	Q4_K_M · 19.6 GB

#	Model	Speed	Quant · Memory
1	Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 49.3 t/s	49.3 tok/s	Q4_K_M · 17.9 GB
2	Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 42.2 t/s	42.2 tok/s	Q4_K_M · 21.7 GB
3	Qwen3.6 35B A3BAlibaba Q4_K_M · 22.0 GB · 87.9 t/s	87.9 tok/s	Q4_K_M · 22.0 GB
4	Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 109 t/s	109 tok/s	Q4_K_M · 16.5 GB
5	Muse Glimmer 30BMeta Q4_K_M · 19.6 GB · 46.9 t/s	46.9 tok/s	Q4_K_M · 19.6 GB

#	Model	Speed	Quant · Memory
1	Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 49.3 t/s	49.3 tok/s	Q4_K_M · 17.9 GB
2	Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 42.2 t/s	42.2 tok/s	Q4_K_M · 21.7 GB
3	Qwen3.5 27BAlibaba Q4_K_M · 17.9 GB · 49.3 t/s	49.3 tok/s	Q4_K_M · 17.9 GB
4	Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 109 t/s	109 tok/s	Q4_K_M · 16.5 GB
5	Muse Glimmer 30BMeta Q4_K_M · 19.6 GB · 46.9 t/s	46.9 tok/s	Q4_K_M · 19.6 GB

#	Model	Speed	Math Score	Quant · Memory
1	Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 49.3 t/s	49.3 tok/s		Q4_K_M · 17.9 GB
2	Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 49.3 t/s	49.3 tok/s		Q4_K_M · 17.9 GB
3	Qwen3.6 35B A3BAlibaba Q4_K_M · 22.0 GB · 87.9 t/s	87.9 tok/s		Q4_K_M · 22.0 GB
4	Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 42.2 t/s	42.2 tok/s	overall	Q4_K_M · 21.7 GB
5	Muse Glimmer 30BMeta Q4_K_M · 19.6 GB · 46.9 t/s	46.9 tok/s		Q4_K_M · 19.6 GB

#	Model	Speed ↓	Quant · Memory
1	Qwen3.5 4BAlibabaHow to run → Q4_K_M · 3.4 GB · 264 t/s	264 tok/s	Q4_K_M · 3.4 GB
2	Qwen3.5 9BAlibaba Q4_K_M · 6.6 GB · 135 t/s	135 tok/s	Q4_K_M · 6.6 GB
3	GPT OSS 20BOpenAI Q4_K_M · 13.1 GB · 128 t/s	128 tok/s	Q4_K_M · 13.1 GB
4	Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 109 t/s	109 tok/s	Q4_K_M · 16.5 GB
5	Muse Glimmer 30BMeta Q4_K_M · 19.6 GB · 46.9 t/s	46.9 tok/s	Q4_K_M · 19.6 GB

See all 390 models on the Mac Studio M5 Ultra →Bars = traction for models without a Score yet

About the Mac Studio M5 Ultra (96 GB)

The Mac Studio M5 Ultra is the fastest 96 GB machine: it runs 70–120B models well. macOS keeps about 3.5 GB, leaving 92.5 GB for models. With 109 TFLOPS it reads long prompts faster than most 96 GB machines: a 4K-token prompt takes 10 s. At its $5,499 MSRP it gives 16.0 tok/s per $1,000, 4th of the 7 machines compared here. M5 Ultra generation, released 2026.

Speed in 96 GB#11.2 TB/s

Prompt reading#1109 TFLOPS

For models92.5 GBmacOS keeps 3.5 GB

ChipM5 Ultrareleased 2026

Mac Studio M5 Ultra strengths and weaknesses

Strengths

Runs 120B MoE models: GPT OSS 120B at 89.0 tok/s in 70.8 GB
Room for long context: Llama 3.3 70B Instruct at Q4 uses 51.3 GB, leaving 41.2 GB free

Weaknesses

Slow prompt reading: A 32K-token prompt on Qwen3.8 27B waits 1.4 min for the first token
No CUDA: vLLM, AWQ, GPTQ and FP8 builds don't run; use GGUF or MLX

Memory for models

92.5 GB of 96 GB

35B MoE22 GB70B51.3 GB120B MoE70.8 GB127B82.8 GB

35B MoE22 GB70B51.3 GB127B82.8 GB

For models 92.5 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 56.8 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 M5 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 $5,502	213	38.8	37.1
Framework Desktop Ryzen AI Max+ 395128 GB	about $1,999	46.7	23.4	33.4
Mac Studio M3 Ultra96 GB	about $3,999	76.6	19.2	28.4
Mac Studio M5 Ultra96 GB	about $5,499	87.9	16.0	–
Mac Studio M2 Ultra192 GB	about $5,599	75.9	13.6	25.7
Mac Studio M5 Ultra256 GB	about $9,499	87.9	9.3	–
Mac Studio M5 Ultra512 GB	about $22,299	87.9	3.9	–

Machine	Used price	tok/s	tok/s per $1,000	tok/s per 100 W
RTX 409024 GBBest used value	about $3,000	151	50.3	33.6
RTX 509032 GB	about $7,000	213	30.5	37.1
Mac Studio M3 Ultra96 GB	about $4,525	76.6	16.9	28.4

Cost per token: the Mac Studio M5 Ultra against 24–512 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.

NewUsed

Speed$ per 1M tokens

24–36 GB96–128 GB192+ GBthis desktopat MSRP

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

$3.97

Machine, MSRP: $3.97
Power: not published

3-year cost of ownership$5,499

Or rent an RTX PRO 6000 Blackwell Max-Q Workstation Edition in the cloud$0.76/hr

Renting costs as much as buying after ≈7,200 hours (≈4.9 years at 4 h a day).

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. Machines without a published power figure are left out of Per watt and costed without power. 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, 3× the memory for $4,000 moreMac Studio M5 Ultra (256 GB)Apple · 252.5 GB for models · M5 Ultra

same speed▲ $4,000 more▲ 160 GB$9,499

Same family Same chip, 5× the memory for $16,800 moreMac Studio M5 Ultra (512 GB)Apple · 508.5 GB for models · M5 Ultra

same speed▲ $16,800 more▲ 416 GB$22,299

Same family Machine, 13% slowerMac Studio M3 Ultra (96 GB)Apple · 92.5 GB for models · M3 Ultra

▼ 13% speedsame memory$3,999

Machine, 47% slowerFramework Desktop Ryzen AI Max+ 395Framework · 96 GB for models · Ryzen AI Max+ 395

▼ 47% speed▲ 3.5 GB$1,999

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

▲ 72% speed▼ 68.5 GB$1,599rent $0.34/hr

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

▲ 143% speed▼ 60.5 GB$5,502rent $0.50/hr

Mac Studio M5 Ultra tokens per second by model size

Output speed depends on memory bandwidth (1228 GB/s). Grey rows need more than the 92.5 GB available.

9BQwen3.5 9B · Q4_K_M · 6.6 GB
135tok/s
14BQwen3 14B · Q4_K_M · 11.9 GB
90.5tok/s
27–31BQwen3.8 27B · Q4_K_M · 17.9 GB
49.3tok/s
35B MoEQwen3.6 35B A3B · Q4_K_M · 22.0 GB
87.9tok/s
70BLlama 3.3 70B Instruct · Q4_K_M · 51.3 GB
18.5tok/s
120B MoEGPT OSS 120B · Q4_K_M · 70.8 GB
89.0tok/s

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

Ultra generations: writing vs compute

M1 Ultra → M5 Ultra

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

Writing speed follows memory bandwidth and barely moved: 75.9 to 87.9 tok/s since the M1 Ultra. Compute, which sets how fast prompts are read, grew 3.5×. Indexed to the M1 Ultra; speeds on Qwen3.6 35B A3B, same 128 GB.

All 1280 models on the Mac Studio M5 Ultra

Featured only

Best quant that fits in 92.5 GB, ordered by llmrun Score. Fit grades how well it runs here, S best. How ranking works

LLM models compatible with Mac Studio M5 Ultra (96 GB) — ranked by Score, recency and Fit
Model	Quant	VRAM	Speed	Context	Fit
Nanbeige4.2 3B4.2BChatFunctions Q4_K_M·297.4 t/s	Q4_K_M	3.5 GB4%	297.4 t/s	262K	S100, Runs great
Llama 4 Scout 17B 16E Instruct108.6BVision Q4_K_M·44.1 t/s	Q4_K_M	77.0 GB83%	44.1 t/s	—	A73, Runs well
Wald 4B4.7BChatReasoningFunctions Q4_K_M·263.7 t/s	Q4_K_M	3.4 GB4%	263.7 t/s	262K	S100, Runs great
Crowd V11.0BChatFunctions BF16·359.7 t/s	BF16	2.4 GB3%	359.7 t/s	2K	S100, Runs great
Supra2 100M Instruct101MChat Q4_K_M·2204.1 t/s	Q4_K_M	0.4 GB0%	2204.1 t/s	2K	S100, Runs great
AREX 227.4BVisionFunctionsReasoning Q4_K_M·50.1 t/s	Q4_K_M	17.6 GB19%	50.1 t/s	262K	A79, Runs well
Meta Llama 3.1 8B Instruct8.0BChat Q4_K_M·162.2 t/s	Q4_K_M	7.9 GB9%	162.2 t/s	—	S97, Runs great
Agents A1 4B4.5BChatVisionFunctions Q4_K_M·269.5 t/s	Q4_K_M	3.4 GB4%	269.5 t/s	262K	S100, Runs great
Sevclassifier Jev Style 1.5B1.5BChatFunctions Q4_K_M·671.6 t/s	Q4_K_M	1.7 GB2%	671.6 t/s	33K	S100, Runs great
Holo4 27B27.4BVisionFunctions Q4_K_M·50.1 t/s	Q4_K_M	17.6 GB19%	50.1 t/s	262K	A79, Runs well
Inkling Small DSpark Preview1.4BChat Q4_K_M·728.5 t/s	Q4_K_M	1.5 GB2%	728.5 t/s	1049K	S100, Runs great
Cosmos Reason2 8B8.8BVisionChatReasoningFunctions Q4_K_M·148.5 t/s	Q4_K_M	8.4 GB9%	148.5 t/s	—	S97, Runs great
Hy MT2 30B A3B30.1BTranslationFunctions Q4_K_M·82.9 t/s	Q4_K_M	19.1 GB21%	82.9 t/s	262K	S88, Runs great
Deepseek V4 Flash 0731 Spark60.3BChat Q4_K_M·23.5 t/s	Q4_K_M	36.7 GB40%	23.5 t/s	1049K	B64, Decent
Edge0 35B A3B Preview34.7BChatFunctions Q4_K_M·95.6 t/s	Q4_K_M	21.3 GB23%	95.6 t/s	262K	S90, Runs great

Mac Studio M5 Ultra specs

Specs

Chip: M5 Ultra
CPU cores: 30
GPU cores: 64
Unified memory: 96 GB
For models: 92.5 GB (macOS keeps 3.5 GB)
Memory bandwidth: 1.2 TB/s
FP16 (dense): 108.8 TFLOPS
Neural Engine: 76 TOPS
Released: 2026-09-22
MSRP: $5,499

Mac Studio M5 Ultra FAQ

Is the Mac Studio M5 Ultra (96 GB) worth it for local LLMs in 2026?

At its $5,499 MSRP it gives 16.0 tok/s per $1,000 on Qwen3.6 35B A3B, 4th of the 7 machines compared here. It runs 70–120B models well.

How much memory is available for AI on the Mac Studio M5 Ultra (96 GB)?

About 92.5 GB of the 96 GB. macOS keeps roughly 3.5 GB for itself and other apps.

What is the largest model the Mac Studio M5 Ultra (96 GB) can run?

Mistral Medium 3.5 128B (127.7B) is the largest at 4-bit or better: at Q4_K_M it uses 82.8 GB, 90% of the available memory, and runs at about 11.1 tok/s.

Can the Mac Studio M5 Ultra (96 GB) run Qwen3.8 27B?

Yes. Qwen3.8 27B fits at Q4_K_M, using 17.9 GB of the 92.5 GB available, and it is the best model that runs well on this machine.

How long does the Mac Studio M5 Ultra (96 GB) take to read a long prompt?

On Qwen3.8 27B, about 10 s for a 4K-token prompt and 84 s for a 32K-token document before the first word appears.

Is renting a GPU cheaper than buying the Mac Studio M5 Ultra?

An RTX PRO 6000 Blackwell Max-Q Workstation Edition (96 GB) rents from $0.76/hr. Renting costs as much as buying after ≈7,200 hours (≈4.9 years at 4 h a day).

48 GB GPUs All Macs Desktops & mini PCs Mac Pro M2 Ultra Mac Studio M2 Max MacBook Pro 14" M2 Max MacBook Pro 16" M2 Max Mac Studio M4 Max Best coding LLMs All hardware