readygpu

Beelink GTR9 Pro Ryzen AI Max+ 395 128GB vs Mac Studio M4 Max 128GB

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

In short

Both have 128 GB of memory for models; the Beelink GTR9 Pro Ryzen AI Max+ 395 128GB has 256 GB/s of bandwidth and the Mac Studio M4 Max 128GB 546 GB/s. At 32k context both run the same 72 models we track fully in memory. On the models both run fully in memory, the Mac Studio M4 Max 128GB is about 2.5× as fast as the Beelink GTR9 Pro Ryzen AI Max+ 395 128GB (estimated, median across models).

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

Beelink GTR9 Pro Ryzen AI Max+ 395 128GB

Not priced here

Add to build
Mac Studio M4 Max 128GB

Not priced here

Add to build

Specifications

SpecBeelink GTR9 Pro Ryzen AI Max+ 395 128GBMac Studio M4 Max 128GB
ChipRyzen AI Max+ 395 (Radeon 8060S)Apple M4 Max (40-core GPU)
Memory bus256-bit512-bit
Memory bandwidth256 GB/s546 GB/s
FP16 compute59.4 TFLOPS16.16 TFLOPS
SoftwareROCmMetal
Unified memory128 GB128 GB
GPU can use by default96 GB96 GB
GPU can use, tuned120 GB120 GB
Operating systemwindowsmacos
ProcessorRyzen AI Max+ 395M4 Max 16-core
Storage2000 GB1000 GB

What each runs

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

ModelBeelink GTR9 Pro Ryzen AI Max+ 395 128GBMac Studio M4 Max 128GB
DeepSeek V4 Flash9tok/sUD-Q2_K_XL22tok/sUD-Q2_K_XL
Qwen 3.8 Flash Next3tok/sUD-Q3_K_XL7tok/sUD-Q3_K_XL
MiniMax M2.77tok/sUD-Q2_K_XL18tok/sUD-Q2_K_XL
Qwen 3.8 27B5tok/sQ8_011tok/sQ8_0
Qwen 3.5 122B-A10B9tok/sQ6_K24tok/sQ6_K
Qwen 3.6 27B5tok/sQ8_012tok/sQ8_0
gpt-oss 120B20tok/sMXFP449tok/sMXFP4
Qwen3 Coder Next17tok/sQ8_043tok/sQ8_0
Nemotron 3 Super 120B-A12B9tok/sQ5_K_M22tok/sQ5_K_M
Gemma 4 31B4tok/sQ8_010tok/sQ8_0
Mistral Small 4 119B13tok/sUD-Q6_K32tok/sUD-Q6_K
Qwen 3.6 35B-A3B22tok/sQ8_056tok/sQ8_0
Qwen 3.5 27B5tok/sQ8_012tok/sQ8_0
Gemma 4 26B-A4B17tok/sQ8_042tok/sQ8_0
GLM 4.5 Air4tok/sQ5_K_M11tok/sQ5_K_M
Qwen 3.5 35B-A3B22tok/sQ8_056tok/sQ8_0
Qwen3 Next 80B-A3B21tok/sQ8_053tok/sQ8_0
GLM 4.7 Flash12tok/sQ8_030tok/sQ8_0
Devstral Small 2 24B5tok/sQ8_012tok/sQ8_0
Nemotron 3.5 Lightning 30B-A3B24tok/sUD-Q6_K_XL60tok/sUD-Q6_K_XL
gpt-oss 20B26tok/sMXFP466tok/sMXFP4
Kimi Linear 48B-A3B24tok/sQ8_061tok/sQ8_0
Qwen3 Coder 30B-A3B13tok/sQ8_033tok/sQ8_0
Qwen3 32B3tok/sQ8_08tok/sQ8_0
Gemma 4 12B10tok/sQ8_026tok/sQ8_0
Qwen 3.5 9B13tok/sQ8_033tok/sQ8_0
Mistral Small 3.2 24B5tok/sQ8_012tok/sQ8_0
Gemma 3 27B4tok/sQ8_011tok/sQ8_0
DeepSeek R1 Distill Qwen 32B3tok/sQ8_08tok/sQ8_0
Llama 3.3 70B2tok/sQ8_04tok/sQ8_0
Ministral 3 14B7tok/sQ8_018tok/sQ8_0
Qwen3 14B7tok/sQ8_017tok/sQ8_0
DeepSeek R1 0528 Qwen3 8B10tok/sQ8_026tok/sQ8_0
Qwen3 8B10tok/sQ8_026tok/sQ8_0
Phi-4 14B6tok/sQ8_016tok/sQ8_0
Qwen 3.5 4B25tok/sQ8_063tok/sQ8_0
Ministral 3 8B10tok/sQ8_026tok/sQ8_0
Granite 4.2 8B10tok/sQ8_024tok/sQ8_0
Gemma 4 E4B15tok/sQ8_039tok/sQ8_0
LFM2.5 8B-A1B32tok/sQ8_080tok/sQ8_0
Llama 3.1 8B13tok/sQ6_K33tok/sQ6_K
Qwen2.5 0.5B151tok/sQ8_0380tok/sQ8_0
Qwen3.5 0.8B114tok/sQ8_0288tok/sQ8_0
Qwen3.5 2B76tok/sQ5_K_M191tok/sQ5_K_M
DeepSeek R1 Distill Qwen 1.5B50tok/sQ8_0125tok/sQ8_0
MiniCPM5 2B34tok/sQ8_087tok/sQ8_0
LFM2.5 2.6B29tok/sQ8_073tok/sQ8_0
LFM2.5 VL 3B29tok/sQ8_073tok/sQ8_0
Mellum2.1 12B A2.5B Thinking27tok/sQ8_067tok/sQ8_0
granite vision 4.1 4b22tok/sQ8_056tok/sQ8_0
granite 4.2 3b21tok/sQ8_054tok/sQ8_0
Kimi VL A3B Thinking 250621tok/sQ8_052tok/sQ8_0
Kimi VL A3B20tok/smlx-8bit50tok/smlx-8bit
LFM2.5 24B A2B18tok/sQ8_044tok/sQ8_0
GLM 4 9B 041415tok/smlx-5bit37tok/smlx-5bit
DeepSeek R1 Distill Qwen 7B14tok/sQ8_036tok/sQ8_0
GLM 4.6V Flash12tok/sQ8_031tok/sQ8_0
DeepSeek R1 Distill Llama 8B11tok/sQ8_028tok/sQ8_0
GLM 4.1V 9B Thinking11tok/smlx-8bit27tok/smlx-8bit
Ministral 3 8B Reasoning 251210tok/sQ8_026tok/sQ8_0
DeepSeek R1 Distill Qwen 14B6tok/sQ8_016tok/sQ8_0
GLM 4.5V6tok/sQ4_K_M15tok/sQ4_K_M
GLM 4.6V5tok/sQ4_K_M14tok/sQ4_K_M
Devstral Small 25075tok/sQ8_012tok/sQ8_0
Magistral Small 25065tok/sQ8_012tok/sQ8_0
Magistral Small 25075tok/sQ8_012tok/sQ8_0
Magistral Small 25095tok/sQ8_012tok/sQ8_0
granite 4.2 30b4tok/sQ8_09tok/sQ8_0
DeepSeek R1 Distill Llama 70B2tok/sQ8_04tok/sQ8_0
Kimi Dev 72B2tok/sQ8_04tok/sQ8_0
Devstral 2 123B Instruct 25121tok/sQ5_K_M4tok/sQ5_K_M
Mistral Medium 3.5 128B1tok/sQ5_K_M4tok/sQ5_K_M
GLM 5.3 Flashwon'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 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
GLM 4.6won't runwon't run
GLM 4.7won'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

Questions

Which is better for local AI, the Beelink GTR9 Pro Ryzen AI Max+ 395 128GB or the Mac Studio M4 Max 128GB?
Choose the Mac Studio M4 Max 128GB for speed.
Which can run bigger models, the Beelink GTR9 Pro Ryzen AI Max+ 395 128GB or the Mac Studio M4 Max 128GB?
At 32k context both run the same 72 models we track fully in memory.
Which is faster for local AI, the Beelink GTR9 Pro Ryzen AI Max+ 395 128GB or the Mac Studio M4 Max 128GB?
On the models both run fully in memory, the Mac Studio M4 Max 128GB is about 2.5× as fast as the Beelink GTR9 Pro Ryzen AI Max+ 395 128GB (estimated, median across models).

More comparisons