readygpu

Mac Studio M3 Ultra 256GB vs Mac Studio M3 Ultra 96GB

For running models locally, priced in the US. At 32k context the first runs 42 models fully in memory, the second 38.

In short

The Mac Studio M3 Ultra 256GB has more memory for models: 256 GB against 96 GB. At 32k context the Mac Studio M3 Ultra 256GB runs 42 of the models we track fully in memory and the Mac Studio M3 Ultra 96GB 38; only the Mac Studio M3 Ultra 256GB holds DeepSeek V4 Flash. On the models both run fully in memory, the Mac Studio M3 Ultra 256GB and the Mac Studio M3 Ultra 96GB are about as fast as each other (estimated).

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

Mac Studio M3 Ultra 256GB

Not priced here

Add to build
Mac Studio M3 Ultra 96GB

Not priced here

Add to build

Specifications

SpecMac Studio M3 Ultra 256GBMac Studio M3 Ultra 96GB
ChipApple M3 Ultra (80-core GPU)Apple M3 Ultra (80-core GPU)
Memory bus1024-bit1024-bit
Memory bandwidth819 GB/s819 GB/s
FP16 compute28 TFLOPS28 TFLOPS
SoftwareMetalMetal
Unified memory256 GB96 GB
GPU can use by default192 GB72 GB
GPU can use, tuned248 GB88 GB
Operating systemmacosmacos
ProcessorM3 Ultra 32-coreM3 Ultra 32-core
Storage2000 GB1000 GB

What each runs

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

ModelMac Studio M3 Ultra 256GBMac Studio M3 Ultra 96GB
DeepSeek V4 Flash23tok/sUD-Q4_K_XLwon't run
GLM 5.3 Flash20tok/sUD-Q4_K_XLwon't run
Qwen 3.8 Flash Next5tok/sQ8_0won't run
MiniMax M2.718tok/sUD-Q6_Kwon't run
Qwen 3.8 27B17tok/sQ8_017tok/sQ8_0
Qwen 3.5 122B-A10B28tok/sQ8_058tok/sQ3_K_M
Qwen 3.6 27B17tok/sQ8_017tok/sQ8_0
gpt-oss 120B74tok/sMXFP474tok/sMXFP4
Qwen3 Coder Next64tok/sQ8_078tok/sQ6_K
Nemotron 3 Super 120B-A12B25tok/sQ8_050tok/sUD-Q3_K_M
Gemma 4 31B15tok/sQ8_015tok/sQ8_0
Mistral Small 4 119B39tok/sQ8_062tok/sUD-Q4_K_M
Qwen 3.6 35B-A3B84tok/sQ8_084tok/sQ8_0
Qwen 3.5 27B17tok/sQ8_017tok/sQ8_0
Gemma 4 26B-A4B63tok/sQ8_063tok/sQ8_0
GLM 4.5 Air13tok/sQ8_021tok/sQ3_K_M
Qwen 3.5 35B-A3B84tok/sQ8_084tok/sQ8_0
Qwen3 Next 80B-A3B80tok/sQ8_097tok/sQ6_K
GLM 4.7 Flash45tok/sQ8_045tok/sQ8_0
Devstral Small 2 24B18tok/sQ8_018tok/sQ8_0
Nemotron 3.5 Lightning 30B-A3B89tok/sUD-Q6_K_XL89tok/sUD-Q6_K_XL
gpt-oss 20B99tok/sMXFP499tok/sMXFP4
Kimi Linear 48B-A3B91tok/sQ8_091tok/sQ8_0
Qwen3 Coder 30B-A3B50tok/sQ8_050tok/sQ8_0
Qwen3 32B12tok/sQ8_012tok/sQ8_0
Gemma 4 12B39tok/sQ8_039tok/sQ8_0
Qwen 3.5 9B50tok/sQ8_050tok/sQ8_0
Mistral Small 3.2 24B18tok/sQ8_018tok/sQ8_0
Gemma 3 27B17tok/sQ8_017tok/sQ8_0
DeepSeek R1 Distill Qwen 32B12tok/sQ8_012tok/sQ8_0
Llama 3.3 70B6tok/sQ8_09tok/sQ5_K_M
Ministral 3 14B27tok/sQ8_027tok/sQ8_0
Qwen3 14B25tok/sQ8_025tok/sQ8_0
DeepSeek R1 0528 Qwen3 8B39tok/sQ8_039tok/sQ8_0
Qwen3 8B39tok/sQ8_039tok/sQ8_0
Phi-4 14B24tok/sQ8_024tok/sQ8_0
Qwen 3.5 4B95tok/sQ8_095tok/sQ8_0
Ministral 3 8B39tok/sQ8_039tok/sQ8_0
Granite 4.2 8B36tok/sQ8_036tok/sQ8_0
Gemma 4 E4B58tok/sQ8_058tok/sQ8_0
LFM2.5 8B-A1B119tok/sQ8_0119tok/sQ8_0
Llama 3.1 8B49tok/sQ6_K49tok/sQ6_K

Questions

Which is better for local AI, the Mac Studio M3 Ultra 256GB or the Mac Studio M3 Ultra 96GB?
Choose the Mac Studio M3 Ultra 256GB to run bigger models.
Which can run bigger models, the Mac Studio M3 Ultra 256GB or the Mac Studio M3 Ultra 96GB?
At 32k context the Mac Studio M3 Ultra 256GB runs 42 of the models we track fully in memory and the Mac Studio M3 Ultra 96GB 38; only the Mac Studio M3 Ultra 256GB holds DeepSeek V4 Flash.
Which is faster for local AI, the Mac Studio M3 Ultra 256GB or the Mac Studio M3 Ultra 96GB?
On the models both run fully in memory, the Mac Studio M3 Ultra 256GB and the Mac Studio M3 Ultra 96GB are about as fast as each other (estimated).

More comparisons