readygpu

Mac Studio M3 Ultra 96GB vs Mac Studio M4 Max 64GB

For running models locally, priced in the UK. At 32k context the first runs 38 models fully in memory, the second 34.

In short

The Mac Studio M3 Ultra 96GB has more memory for models: 96 GB against 64 GB. At 32k context the Mac Studio M3 Ultra 96GB runs 38 of the models we track fully in memory and the Mac Studio M4 Max 64GB 34; only the Mac Studio M3 Ultra 96GB holds gpt-oss 120B. On the models both run fully in memory, the Mac Studio M3 Ultra 96GB is about 1.5× as fast as the Mac Studio M4 Max 64GB (estimated, median across models).

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

Mac Studio M3 Ultra 96GB

Not priced here

Add to build
Mac Studio M4 Max 64GB

Not priced here

Add to build

Specifications

SpecMac Studio M3 Ultra 96GBMac Studio M4 Max 64GB
ChipApple M3 Ultra (80-core GPU)Apple M4 Max (40-core GPU)
Memory bus1024-bit512-bit
Memory bandwidth819 GB/s546 GB/s
FP16 compute28 TFLOPS16.16 TFLOPS
SoftwareMetalMetal
Unified memory96 GB64 GB
GPU can use by default72 GB48 GB
GPU can use, tuned88 GB56 GB
Operating systemmacosmacos
ProcessorM3 Ultra 32-coreM4 Max 16-core
Storage1000 GB1000 GB

What each runs

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

ModelMac Studio M3 Ultra 96GBMac Studio M4 Max 64GB
Qwen 3.8 27B17tok/sQ8_011tok/sQ8_0
Qwen 3.5 122B-A10B58tok/sQ3_K_M50tok/sUD-Q2_K_XL
Qwen 3.6 27B17tok/sQ8_012tok/sQ8_0
Qwen3 Coder Next78tok/sQ6_K66tok/sQ4_K_M
gpt-oss 120B74tok/sMXFP4won't run
Nemotron 3 Super 120B-A12B50tok/sUD-Q3_K_Mwon't run
Gemma 4 31B15tok/sQ8_010tok/sQ8_0
Mistral Small 4 119B62tok/sUD-Q4_K_M68tok/sUD-Q2_K_XL
Qwen 3.6 35B-A3B84tok/sQ8_056tok/sQ8_0
Qwen 3.5 27B17tok/sQ8_012tok/sQ8_0
Gemma 4 26B-A4B63tok/sQ8_042tok/sQ8_0
GLM 4.5 Air21tok/sQ3_K_Mwon't run
Qwen3 Next 80B-A3B97tok/sQ6_K80tok/sQ4_K_M
Qwen 3.5 35B-A3B84tok/sQ8_056tok/sQ8_0
GLM 4.7 Flash45tok/sQ8_030tok/sQ8_0
Devstral Small 2 24B18tok/sQ8_012tok/sQ8_0
Nemotron 3.5 Lightning 30B-A3B89tok/sUD-Q6_K_XL60tok/sUD-Q6_K_XL
gpt-oss 20B99tok/sMXFP466tok/sMXFP4
Kimi Linear 48B-A3B91tok/sQ8_077tok/sQ6_K
Qwen3 Coder 30B-A3B50tok/sQ8_033tok/sQ8_0
Qwen3 32B12tok/sQ8_08tok/sQ8_0
Gemma 4 12B39tok/sQ8_026tok/sQ8_0
Qwen 3.5 9B50tok/sQ8_033tok/sQ8_0
Mistral Small 3.2 24B18tok/sQ8_012tok/sQ8_0
Gemma 3 27B17tok/sQ8_011tok/sQ8_0
DeepSeek R1 Distill Qwen 32B12tok/sQ8_08tok/sQ8_0
Llama 3.3 70B9tok/sQ5_K_Mwon't run
Ministral 3 14B27tok/sQ8_018tok/sQ8_0
Qwen3 14B25tok/sQ8_017tok/sQ8_0
DeepSeek R1 0528 Qwen3 8B39tok/sQ8_026tok/sQ8_0
Qwen3 8B39tok/sQ8_026tok/sQ8_0
Phi-4 14B24tok/sQ8_016tok/sQ8_0
Qwen 3.5 4B95tok/sQ8_063tok/sQ8_0
Ministral 3 8B39tok/sQ8_026tok/sQ8_0
Granite 4.2 8B36tok/sQ8_024tok/sQ8_0
Gemma 4 E4B58tok/sQ8_039tok/sQ8_0
LFM2.5 8B-A1B119tok/sQ8_080tok/sQ8_0
Llama 3.1 8B49tok/sQ6_K33tok/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

Questions

Which is better for local AI, the Mac Studio M3 Ultra 96GB or the Mac Studio M4 Max 64GB?
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 Studio M3 Ultra 96GB or the Mac Studio M4 Max 64GB?
At 32k context the Mac Studio M3 Ultra 96GB runs 38 of the models we track fully in memory and the Mac Studio M4 Max 64GB 34; only the Mac Studio M3 Ultra 96GB holds gpt-oss 120B.
Which is faster for local AI, the Mac Studio M3 Ultra 96GB or the Mac Studio M4 Max 64GB?
On the models both run fully in memory, the Mac Studio M3 Ultra 96GB is about 1.5× as fast as the Mac Studio M4 Max 64GB (estimated, median across models).

More comparisons