readygpu

Beelink GTR9 Pro Ryzen AI Max+ 395 128GB vs Mac Studio M3 Ultra 256GB

For running models locally, priced in the US. At 8k context the first runs 78 models fully in memory, the second 84.

In short

The Mac Studio M3 Ultra 256GB has more memory for models: 256 GB against 128 GB. At 8k context the Mac Studio M3 Ultra 256GB runs 84 of the models we track fully in memory and the Beelink GTR9 Pro Ryzen AI Max+ 395 128GB 78; only the Mac Studio M3 Ultra 256GB holds GLM 5.3 Flash. On the models both run fully in memory, the Mac Studio M3 Ultra 256GB is about 3.8× 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 M3 Ultra 256GB

$7,999

Amazon

Add to build

Specifications

SpecBeelink GTR9 Pro Ryzen AI Max+ 395 128GBMac Studio M3 Ultra 256GB
ChipRyzen AI Max+ 395 (Radeon 8060S)Apple M3 Ultra (80-core GPU)
Memory bus256-bit1024-bit
Memory bandwidth256 GB/s819 GB/s
FP16 compute59.4 TFLOPS28 TFLOPS
SoftwareROCmMetal
Unified memory128 GB256 GB
GPU can use by default96 GB192 GB
GPU can use, tuned120 GB248 GB
Operating systemwindowsmacos
ProcessorRyzen AI Max+ 395M3 Ultra 32-core
Storage2000 GB2000 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 M3 Ultra 256GB
DeepSeek V4 Flash11tok/sUD-Q2_K_XL27tok/sMXFP4
Qwen 3.8 Flash Next3tok/sUD-Q3_K_XL5tok/sQ8_0
GLM 5.3 Flashwon't run20tok/sUD-Q4_K_XL
MiniMax M2.715tok/sUD-Q2_K_XL28tok/sQ6_K
Qwen 3.8 27B5tok/sQ8_018tok/sQ8_0
Qwen 3.5 122B-A10B10tok/sQ6_K30tok/sQ8_0
Qwen 3.6 27B5tok/sQ8_018tok/sQ8_0
gpt-oss 120B25tok/sMXFP493tok/sMXFP4
Qwen3 Coder Next19tok/sQ8_073tok/sQ8_0
Nemotron 3 Super 120B-A12B9tok/sQ5_K_M25tok/sQ8_0
Gemma 4 31B4tok/sQ8_016tok/sQ8_0
Mistral Small 4 119B14tok/sUD-Q6_K42tok/sQ8_0
Qwen 3.6 35B-A3B26tok/sQ8_097tok/sQ8_0
Qwen 3.5 27B5tok/sQ8_018tok/sQ8_0
Gemma 4 26B-A4B19tok/sQ8_070tok/sQ8_0
GLM 4.5 Air6tok/sQ6_K20tok/sQ8_0
Qwen 3.5 35B-A3B26tok/sQ8_097tok/sQ8_0
Qwen3 Next 80B-A3B25tok/sQ8_094tok/sQ8_0
GLM 4.7 Flash15tok/sQ8_056tok/sQ8_0
Devstral Small 2 24B5tok/sQ8_020tok/sQ8_0
Nemotron 3.5 Lightning 30B-A3B25tok/sUD-Q6_K_XL93tok/sUD-Q6_K_XL
gpt-oss 20B32tok/sMXFP4121tok/sMXFP4
Kimi Linear 48B-A3B26tok/sQ8_097tok/sQ8_0
Qwen3 Coder 30B-A3B21tok/sQ8_080tok/sQ8_0
Qwen3 32B4tok/sQ8_014tok/sQ8_0
Gemma 4 12B11tok/sQ8_041tok/sQ8_0
Qwen 3.5 9B14tok/sQ8_054tok/sQ8_0
Mistral Small 3.2 24B5tok/sQ8_020tok/sQ8_0
Gemma 3 27B5tok/sQ8_018tok/sQ8_0
DeepSeek R1 Distill Qwen 32B4tok/sQ8_014tok/sQ8_0
Llama 3.3 70B2tok/sQ8_07tok/sQ8_0
Ministral 3 14B9tok/sQ8_034tok/sQ8_0
Qwen3 14B8tok/sQ8_031tok/sQ8_0
DeepSeek R1 0528 Qwen3 8B14tok/sQ8_054tok/sQ8_0
Qwen3 8B14tok/sQ8_054tok/sQ8_0
Phi-4 14B8tok/sQ8_031tok/sQ8_0
Qwen 3.5 4B29tok/sQ8_0111tok/sQ8_0
Ministral 3 8B14tok/sQ8_052tok/sQ8_0
Granite 4.2 8B13tok/sQ8_050tok/sQ8_0
Gemma 4 E4B17tok/sQ8_063tok/sQ8_0
LFM2.5 8B-A1B58tok/sQ8_0218tok/sQ8_0
Llama 3.1 8B18tok/sQ6_K69tok/sQ6_K
Qwen2.5 0.5B223tok/sQ8_0843tok/sQ8_0
Qwen3.5 0.8B151tok/sQ8_0572tok/sQ8_0
Qwen3.5 2B91tok/sQ5_K_M342tok/sQ5_K_M
DeepSeek R1 Distill Qwen 1.5B66tok/sQ8_0250tok/sQ8_0
MiniCPM5 2B46tok/sQ8_0176tok/sQ8_0
LFM2.5 2.6B42tok/sQ8_0158tok/sQ8_0
LFM2.5 VL 3B42tok/sQ8_0158tok/sQ8_0
Spark X2.5 4B41tok/smlx-4bit157tok/smlx-4bit
granite vision 4.1 4b33tok/sQ8_0124tok/sQ8_0
Ministral 3 3B Instruct 251231tok/sQ8_0118tok/sQ8_0
granite 4.2 3b31tok/sQ8_0117tok/sQ8_0
LFM2.5 24B A2B30tok/sQ8_0114tok/sQ8_0
Mellum2.1 12B A2.5B Thinking30tok/sQ8_0113tok/sQ8_0
FrogNano 4B 260929tok/sQ8_0108tok/sQ8_0
Kimi VL A3B Thinking 250625tok/sQ8_095tok/sQ8_0
Kimi VL A3B24tok/smlx-8bit91tok/smlx-8bit
DeepSeek R1 Distill Qwen 7B16tok/sQ8_062tok/sQ8_0
GLM 4 9B 041416tok/smlx-5bit62tok/smlx-5bit
DeepSeek R1 Distill Llama 8B15tok/sQ8_055tok/sQ8_0
Ministral 3 8B Reasoning 251214tok/sQ8_052tok/sQ8_0
GLM 4.6V Flash14tok/sQ8_052tok/sQ8_0
GLM 4.1V 9B Thinking12tok/smlx-8bit44tok/smlx-8bit
Ministral 3 14B Reasoning 25129tok/sQ8_034tok/sQ8_0
GLM 4.5V9tok/sQ4_K_M19tok/sQ8_0
DeepSeek R1 Distill Qwen 14B8tok/sQ8_031tok/sQ8_0
GLM 4.6V8tok/sQ4_K_M19tok/sQ8_0
Qwen3.8 27B OBLITERATED7tok/smlx-4bit25tok/smlx-4bit
Devstral Small 25075tok/sQ8_020tok/sQ8_0
Magistral Small 25065tok/sQ8_020tok/sQ8_0
Magistral Small 25075tok/sQ8_020tok/sQ8_0
Magistral Small 25095tok/sQ8_020tok/sQ8_0
Muse Glimmer 30B4tok/sQ8_016tok/sQ8_0
granite 4.2 30b4tok/sQ8_016tok/sQ8_0
DeepSeek R1 Distill Llama 70B2tok/sQ8_07tok/sQ8_0
Kimi Dev 72B2tok/sQ8_07tok/sQ8_0
Devstral 2 123B Instruct 25122tok/sQ5_K_M4tok/sQ8_0
Mistral Medium 3.5 128B2tok/sQ5_K_M4tok/sQ8_0
DeepSeek V4 Flash 0731won't run18tok/sMXFP4
DeepSeek V4 Flash Vision Expwon't run17tok/sMXFP4
MiniMax M2won't run29tok/sQ6_K
MiniMax M2.1won't run29tok/sQ6_K
MiniMax M2.5won't run29tok/sQ6_K
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
GLM 4.5won't runwon't run
GLM 4.6won'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 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 M3 Ultra 256GB?
Choose the Mac Studio M3 Ultra 256GB to run bigger models. Choose the Mac Studio M3 Ultra 256GB for speed.
Which can run bigger models, the Beelink GTR9 Pro Ryzen AI Max+ 395 128GB or the Mac Studio M3 Ultra 256GB?
At 8k context the Mac Studio M3 Ultra 256GB runs 84 of the models we track fully in memory and the Beelink GTR9 Pro Ryzen AI Max+ 395 128GB 78; only the Mac Studio M3 Ultra 256GB holds GLM 5.3 Flash.
Which is faster for local AI, the Beelink GTR9 Pro Ryzen AI Max+ 395 128GB or the Mac Studio M3 Ultra 256GB?
On the models both run fully in memory, the Mac Studio M3 Ultra 256GB is about 3.8× as fast as the Beelink GTR9 Pro Ryzen AI Max+ 395 128GB (estimated, median across models).

More comparisons

More hardware for local AI