readygpu

Mac mini M4 Pro 48GB vs Mac Studio M3 Ultra 96GB

For running models locally, priced in Canada. At 32k context the first runs 61 models fully in memory, the second 72.

In short

The Mac Studio M3 Ultra 96GB has more memory for models: 96 GB against 48 GB. At 32k context the Mac Studio M3 Ultra 96GB runs 72 of the models we track fully in memory and the Mac mini M4 Pro 48GB 61; only the Mac Studio M3 Ultra 96GB holds Qwen 3.5 122B-A10B. On the models both run fully in memory, the Mac Studio M3 Ultra 96GB is about 3.0× as fast as the Mac mini M4 Pro 48GB (estimated, median across models).

Updated . Speeds are estimates; how they are worked out.

Mac mini M4 Pro 48GB

Not priced here

Add to build
Mac Studio M3 Ultra 96GB

Not priced here

Add to build

Specifications

SpecMac mini M4 Pro 48GBMac Studio M3 Ultra 96GB
ChipApple M4 Pro (20-core GPU)Apple M3 Ultra (80-core GPU)
Memory bus256-bit1024-bit
Memory bandwidth273 GB/s819 GB/s
FP16 compute8.08 TFLOPS28 TFLOPS
SoftwareMetalMetal
Unified memory48 GB96 GB
GPU can use by default36 GB72 GB
GPU can use, tuned40 GB88 GB
Operating systemmacosmacos
ProcessorM4 Pro 14-coreM3 Ultra 32-core
Storage512 GB1000 GB

What each runs

Estimated decode speed with the most precise quant that fits each one est.

ModelMac mini M4 Pro 48GBMac Studio M3 Ultra 96GB
Qwen 3.8 27B6tok/sQ8_017tok/sQ8_0
Qwen 3.5 122B-A10Bwon't run58tok/sQ3_K_M
Qwen 3.6 27B6tok/sQ8_017tok/sQ8_0
Gemma 4 31B5tok/sQ8_015tok/sQ8_0
Nemotron 3 Super 120B-A12Bwon't run50tok/sUD-Q3_K_M
Qwen3 Coder Nextwon't run78tok/sQ6_K
gpt-oss 120Bwon't run74tok/sMXFP4
Mistral Small 4 119Bwon't run62tok/sUD-Q4_K_M
Qwen 3.6 35B-A3B28tok/sQ8_084tok/sQ8_0
Qwen 3.5 27B6tok/sQ8_017tok/sQ8_0
Gemma 4 26B-A4B21tok/sQ8_063tok/sQ8_0
GLM 4.5 Airwon't run24tok/sQ3_K_M
Qwen 3.5 35B-A3B28tok/sQ8_084tok/sQ8_0
GLM 4.7 Flash15tok/sQ8_045tok/sQ8_0
Devstral Small 2 24B6tok/sQ8_018tok/sQ8_0
Qwen3 Next 80B-A3Bwon't run97tok/sQ6_K
Nemotron 3.5 Lightning 30B-A3B30tok/sUD-Q6_K_XL89tok/sUD-Q6_K_XL
Kimi Linear 48B-A3B44tok/sQ5_K_M91tok/sQ8_0
gpt-oss 20B33tok/sMXFP499tok/sMXFP4
Qwen3 Coder 30B-A3B17tok/sQ8_050tok/sQ8_0
Qwen3 32B5tok/sQ6_K12tok/sQ8_0
Gemma 4 12B13tok/sQ8_039tok/sQ8_0
Qwen 3.5 9B17tok/sQ8_050tok/sQ8_0
Mistral Small 3.2 24B6tok/sQ8_018tok/sQ8_0
Gemma 3 27B6tok/sQ8_017tok/sQ8_0
DeepSeek R1 Distill Qwen 32B5tok/sQ6_K12tok/sQ8_0
Llama 3.3 70Bwon't run9tok/sQ5_K_M
Ministral 3 14B9tok/sQ8_027tok/sQ8_0
Qwen3 14B8tok/sQ8_025tok/sQ8_0
DeepSeek R1 0528 Qwen3 8B13tok/sQ8_039tok/sQ8_0
Qwen3 8B13tok/sQ8_039tok/sQ8_0
Phi-4 14B8tok/sQ8_024tok/sQ8_0
Qwen 3.5 4B32tok/sQ8_095tok/sQ8_0
Ministral 3 8B13tok/sQ8_039tok/sQ8_0
Granite 4.2 8B12tok/sQ8_036tok/sQ8_0
Gemma 4 E4B19tok/sQ8_058tok/sQ8_0
LFM2.5 8B-A1B40tok/sQ8_0119tok/sQ8_0
Llama 3.1 8B16tok/sQ6_K49tok/sQ6_K
Qwen2.5 0.5B190tok/sQ8_0570tok/sQ8_0
Qwen3.5 0.8B144tok/sQ8_0432tok/sQ8_0
Qwen3.5 2B96tok/sQ5_K_M287tok/sQ5_K_M
DeepSeek R1 Distill Qwen 1.5B63tok/sQ8_0188tok/sQ8_0
MiniCPM5 2B43tok/sQ8_0130tok/sQ8_0
Spark X2.5 4B41tok/smlx-4bit124tok/smlx-4bit
LFM2.5 2.6B36tok/sQ8_0109tok/sQ8_0
LFM2.5 VL 3B36tok/sQ8_0109tok/sQ8_0
Mellum2.1 12B A2.5B Thinking34tok/sQ8_0101tok/sQ8_0
FrogNano 4B 260931tok/sQ8_093tok/sQ8_0
granite vision 4.1 4b28tok/sQ8_084tok/sQ8_0
granite 4.2 3b27tok/sQ8_081tok/sQ8_0
Kimi VL A3B Thinking 250626tok/sQ8_078tok/sQ8_0
Kimi VL A3B25tok/smlx-8bit75tok/smlx-8bit
Ministral 3 3B Instruct 251225tok/sQ8_075tok/sQ8_0
LFM2.5 24B A2B22tok/sQ8_066tok/sQ8_0
GLM 4 9B 041419tok/smlx-5bit56tok/smlx-5bit
DeepSeek R1 Distill Qwen 7B18tok/sQ8_053tok/sQ8_0
GLM 4.6V Flash16tok/sQ8_047tok/sQ8_0
DeepSeek R1 Distill Llama 8B14tok/sQ8_042tok/sQ8_0
GLM 4.1V 9B Thinking14tok/smlx-8bit41tok/smlx-8bit
Ministral 3 8B Reasoning 251213tok/sQ8_039tok/sQ8_0
Ministral 3 14B Reasoning 25129tok/sQ8_027tok/sQ8_0
DeepSeek R1 Distill Qwen 14B8tok/sQ8_024tok/sQ8_0
Qwen3.8 27B OBLITERATED8tok/smlx-4bit23tok/smlx-4bit
Devstral Small 25076tok/sQ8_018tok/sQ8_0
Magistral Small 25066tok/sQ8_018tok/sQ8_0
Magistral Small 25076tok/sQ8_018tok/sQ8_0
Magistral Small 25096tok/sQ8_018tok/sQ8_0
Muse Glimmer 30B5tok/sQ8_016tok/sQ8_0
granite 4.2 30b5tok/sQ6_K13tok/sQ8_0
DeepSeek R1 Distill Llama 70Bwon't run9tok/sQ5_K_M
GLM 4.5Vwon't run22tok/sQ4_K_M
Kimi Dev 72Bwon't run7tok/sQ6_K
DeepSeek V4 Flashwon't runwon't run
GLM 5.3 Flashwon't runwon't run
Qwen 3.8 Flash Nextwon't runwon't run
MiniMax M2.7won't runwon't run
DeepSeek R1won't runwon't run
DeepSeek R1 0528won't runwon't run
DeepSeek R1 Zerowon't runwon't run
DeepSeek V3won't runwon't run
DeepSeek V3 0324won't runwon't run
DeepSeek V3.1won't runwon't run
DeepSeek V3.1 Terminuswon't runwon't run
DeepSeek V3.2won't runwon't run
DeepSeek V4 Flash 0731won't runwon't run
DeepSeek V4 Flash Vision Expwon't runwon't run
Devstral 2 123B Instruct 2512won't runwon't run
GLM 4.5won't runwon't run
GLM 4.6won't runwon't run
GLM 4.6Vwon't runwon't run
GLM 4.7won't runwon't run
GLM 5won't runwon't run
GLM 5.1won't runwon't run
GLM 5.2won't runwon't run
GLM 5.3won't runwon't run
Kimi K2won't runwon't run
Kimi K2 Instruct 0905won't runwon't run
Kimi K2 Thinkingwon't runwon't run
Kimi K2.5won't runwon't run
MiniMax M2won't runwon't run
MiniMax M2.1won't runwon't run
MiniMax M2.5won't runwon't run
MiniMax M3won't runwon't run
Mistral Medium 3.5 128Bwon't runwon't run

Questions

Which is better for local AI, the Mac mini M4 Pro 48GB or the Mac Studio M3 Ultra 96GB?
Choose the Mac Studio M3 Ultra 96GB to run bigger models. Choose the Mac Studio M3 Ultra 96GB for speed.
Which can run bigger models, the Mac mini M4 Pro 48GB or the Mac Studio M3 Ultra 96GB?
At 32k context the Mac Studio M3 Ultra 96GB runs 72 of the models we track fully in memory and the Mac mini M4 Pro 48GB 61; only the Mac Studio M3 Ultra 96GB holds Qwen 3.5 122B-A10B.
Which is faster for local AI, the Mac mini M4 Pro 48GB or the Mac Studio M3 Ultra 96GB?
On the models both run fully in memory, the Mac Studio M3 Ultra 96GB is about 3.0× as fast as the Mac mini M4 Pro 48GB (estimated, median across models).

More comparisons

More hardware for local AI