readygpu

Framework Desktop Ryzen AI Max+ 395 64GB vs Mac mini M4 32GB

For running models locally, priced in the UK. At 32k context the first runs 65 models fully in memory, the second 53.

In short

The Framework Desktop Ryzen AI Max+ 395 64GB has more memory for models: 64 GB against 32 GB. At 32k context the Framework Desktop Ryzen AI Max+ 395 64GB runs 65 of the models we track fully in memory and the Mac mini M4 32GB 53; only the Framework Desktop Ryzen AI Max+ 395 64GB holds Qwen 3.5 122B-A10B. On the models both run fully in memory, the Framework Desktop Ryzen AI Max+ 395 64GB is about 1.8× as fast as the Mac mini M4 32GB (estimated, median across models).

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

Framework Desktop Ryzen AI Max+ 395 64GB

Not priced here

Add to build
Mac mini M4 32GB

£1,599

eBay

Add to build

Specifications

SpecFramework Desktop Ryzen AI Max+ 395 64GBMac mini M4 32GB
ChipRyzen AI Max+ 395 (Radeon 8060S)Apple M4 (10-core GPU)
Memory bus256-bit128-bit
Memory bandwidth256 GB/s120 GB/s
FP16 compute59.4 TFLOPS3.76 TFLOPS
SoftwareROCmMetal
Unified memory64 GB32 GB
GPU can use by default48 GB21.3 GB
GPU can use, tuned56 GB24 GB
Operating systemlinuxmacos
ProcessorRyzen AI Max+ 395M4 10-core
Storage1000 GB512 GB

What each runs

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

ModelFramework Desktop Ryzen AI Max+ 395 64GBMac mini M4 32GB
Qwen 3.8 27B5tok/sQ8_04tok/sQ4_K_M
Qwen 3.5 122B-A10B20tok/sUD-Q2_K_XLwon't run
Qwen 3.6 27B5tok/sQ8_04tok/sQ5_K_M
Qwen3 Coder Next26tok/sQ4_K_Mwon't run
Gemma 4 31B4tok/sQ8_04tok/sQ4_K_M
Mistral Small 4 119B27tok/sUD-Q2_K_XLwon't run
Qwen 3.6 35B-A3B22tok/sQ8_0won't run
Qwen 3.5 27B5tok/sQ8_04tok/sQ5_K_M
Gemma 4 26B-A4B17tok/sQ8_012tok/sQ5_K_M
Qwen3 Next 80B-A3B32tok/sQ4_K_Mwon't run
Qwen 3.5 35B-A3B22tok/sQ8_0won't run
GLM 4.7 Flash12tok/sQ8_010tok/sQ4_K_M
Devstral Small 2 24B5tok/sQ8_04tok/sQ4_K_M
Nemotron 3.5 Lightning 30B-A3B24tok/sUD-Q6_K_XLwon't run
Kimi Linear 48B-A3B30tok/sQ6_Kwon't run
gpt-oss 20B26tok/sMXFP415tok/sMXFP4
Qwen3 Coder 30B-A3B13tok/sQ8_09tok/sQ4_K_M
Qwen3 32B3tok/sQ8_0won't run
Gemma 4 12B10tok/sQ8_06tok/sQ8_0
Qwen 3.5 9B13tok/sQ8_07tok/sQ8_0
Mistral Small 3.2 24B5tok/sQ8_04tok/sQ4_K_M
Gemma 3 27B4tok/sQ8_04tok/sQ4_K_M
DeepSeek R1 Distill Qwen 32B3tok/sQ8_0won't run
Ministral 3 14B7tok/sQ8_04tok/sQ8_0
Qwen3 14B7tok/sQ8_04tok/sQ8_0
DeepSeek R1 0528 Qwen3 8B10tok/sQ8_06tok/sQ8_0
Qwen3 8B10tok/sQ8_06tok/sQ8_0
Phi-4 14B6tok/sQ8_04tok/sQ6_K
Qwen 3.5 4B25tok/sQ8_014tok/sQ8_0
Ministral 3 8B10tok/sQ8_06tok/sQ8_0
Granite 4.2 8B10tok/sQ8_05tok/sQ8_0
Gemma 4 E4B15tok/sQ8_09tok/sQ8_0
LFM2.5 8B-A1B32tok/sQ8_018tok/sQ8_0
Llama 3.1 8B13tok/sQ6_K7tok/sQ6_K
Qwen2.5 0.5B151tok/sQ8_084tok/sQ8_0
Qwen3.5 0.8B114tok/sQ8_063tok/sQ8_0
Qwen3.5 2B76tok/sQ5_K_M42tok/sQ5_K_M
DeepSeek R1 Distill Qwen 1.5B50tok/sQ8_028tok/sQ8_0
MiniCPM5 2B34tok/sQ8_019tok/sQ8_0
Spark X2.5 4B33tok/smlx-4bit18tok/smlx-4bit
LFM2.5 2.6B29tok/sQ8_016tok/sQ8_0
LFM2.5 VL 3B29tok/sQ8_016tok/sQ8_0
Mellum2.1 12B A2.5B Thinking27tok/sQ8_015tok/sQ8_0
FrogNano 4B 260925tok/sQ8_014tok/sQ8_0
granite vision 4.1 4b22tok/sQ8_012tok/sQ8_0
granite 4.2 3b21tok/sQ8_012tok/sQ8_0
Kimi VL A3B Thinking 250621tok/sQ8_011tok/sQ8_0
Kimi VL A3B20tok/smlx-8bit11tok/smlx-8bit
Ministral 3 3B Instruct 251220tok/sQ8_011tok/sQ8_0
LFM2.5 24B A2B18tok/sQ8_011tok/sQ5_K_M
GLM 4 9B 041415tok/smlx-5bit8tok/smlx-5bit
DeepSeek R1 Distill Qwen 7B14tok/sQ8_08tok/sQ8_0
GLM 4.6V Flash12tok/sQ8_07tok/sQ8_0
DeepSeek R1 Distill Llama 8B11tok/sQ8_06tok/sQ8_0
GLM 4.1V 9B Thinking11tok/smlx-8bit6tok/smlx-8bit
Ministral 3 8B Reasoning 251210tok/sQ8_06tok/sQ8_0
Ministral 3 14B Reasoning 25127tok/sQ8_04tok/sQ8_0
DeepSeek R1 Distill Qwen 14B6tok/sQ8_04tok/sQ6_K
Qwen3.8 27B OBLITERATED6tok/smlx-4bitwon't run
Devstral Small 25075tok/sQ8_04tok/sQ4_K_M
Magistral Small 25065tok/sQ8_04tok/sQ4_K_M
Magistral Small 25075tok/sQ8_04tok/sQ4_K_M
Magistral Small 25095tok/sQ8_04tok/sQ4_K_M
Muse Glimmer 30B4tok/sQ8_04tok/sQ5_K_M
granite 4.2 30b4tok/sQ8_0won't run
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
Nemotron 3 Super 120B-A12Bwon't runwon't run
gpt-oss 120Bwon't runwon't run
GLM 4.5 Airwon't runwon't run
Llama 3.3 70Bwon't runwon't run
DeepSeek R1won't runwon't run
DeepSeek R1 0528won't runwon't run
DeepSeek R1 Distill Llama 70Bwon'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 4.5Vwon'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 Dev 72Bwon'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 Framework Desktop Ryzen AI Max+ 395 64GB or the Mac mini M4 32GB?
Choose the Framework Desktop Ryzen AI Max+ 395 64GB to run bigger models. Choose the Framework Desktop Ryzen AI Max+ 395 64GB for speed.
Which can run bigger models, the Framework Desktop Ryzen AI Max+ 395 64GB or the Mac mini M4 32GB?
At 32k context the Framework Desktop Ryzen AI Max+ 395 64GB runs 65 of the models we track fully in memory and the Mac mini M4 32GB 53; only the Framework Desktop Ryzen AI Max+ 395 64GB holds Qwen 3.5 122B-A10B.
Which is faster for local AI, the Framework Desktop Ryzen AI Max+ 395 64GB or the Mac mini M4 32GB?
On the models both run fully in memory, the Framework Desktop Ryzen AI Max+ 395 64GB is about 1.8× as fast as the Mac mini M4 32GB (estimated, median across models).

More comparisons

More hardware for local AI