Machines / Ryzen AI Max+ 395 mini PC (128GB)

What LLMs can the Ryzen AI Max+ 395 mini PC (128GB) run?

The Ryzen AI Max+ 395 mini PC (128GB) has 128 GB of memory (about 96 GB usable by a model) and 256 GB/s of bandwidth. The largest model it runs at a comfortable 10+ tokens/sec is Nemotron 3 super 120B (A12B).

128 GBmemory

256 GB/sbandwidth

Check the Ryzen AI Max+ 395 mini PC (128GB) price →

Fast (20+ tok/s)

Qwen3.5 0.8B 120+DeepSeek-R1 1.5B 120+Qwen 1.8B 120+Llama 3.2 1B 120+Qwen 3 1.7B 120+Granite 3.1 moe 1B 120+lfm 2.5 8B (A1B) 120+minicpm v4.6 1B 112 Gemma 2 2B 112 codeGemma 2B 112 Gemma 2B 105 smollm 2 1.7B 100 starCoder 2 3B 100 falcon 3 1B 100 Qwen2.5 3B 94 Qwen2.5 Coder 3B 94 Qwen3.5 2B 94 Qwen 3-VL 2B 94 Llama 3.2 3B 90 orca mini 3B 90 Granite 3.1 moe 3B 90 falcon 3 3B 90 Granite 4.1 3B 85 Granite 4.2 3B 82 Phi 3 3.8B 82 Qwen 4B 78 Qwen 3 4B 69 Qwen2.5-VL 3B 56 Gemma 3 4B 54 Qwen 3-VL 4B 54 translateGemma 4B 54 Qwen3.5 4B 53 Qwen 3 30B (A3B) 52 Qwen3 Coder 30B (A3B) 52 Qwen 3-VL 30B (A3B) 50 Llama 2 7B 47 Code Llama 7B 47 DeepSeek Coder 6.7B 47 Llama 2 uncensored 7B 47 Qwen3.5 35B (A3B) 47 Qwen3.6 35B (A3B) 47 orca mini 7B 44 gpt-oss 20B 44 Mistral 7B 41 Qwen2 7B 41 starCoder 2 7B 41 OLMo 2 7B 40 Nemotron 3.5 lightning 30B (A3B) 40 falcon 3 7B 39 DeepSeek-R1 7B 38 Qwen2.5 7B 38 Llama 3 8B 38 Qwen2.5 Coder 7B 38 LLaVA 7B 38 Llama 3.1 8B 37 Qwen 7B 37 dolPhin 3 8B 37 Gemma 7B 36 codeGemma 7B 36 Gemma 4 26B (A4B) 36 DeepSeek-R1 8B 35 Qwen 3 8B 35 Granite 4.2 8B 34 Granite 4.1 8B 34 Gemma 2 9B 33 minicpm v 8B 33 LLaVA Llama3 8B 33 ornith 9B 32 Gemma 3n E2B 32 gpt-oss 120B 31 Qwen2.5-VL 7B 30 minicpm v4.5 8B 29 Qwen 3-VL 8B 29 falcon 3 10B 28 ornith 1.5 9B 27 Qwen3.5 9B 27 Mistral nemo 12B 25 Gemma 4 E2B 25 Llama 2 13B 24 Code Llama 13B 24 Gemma 3n E4B 24 Gemma 4 12B 24 Llama 3.2 Vision 11B 23 Phi 3 14B 23 orca mini 13B 23 LLaVA 13B 22 Gemma 3 12B 22 translateGemma 12B 22 Qwen 14B 22 OLMo 2 13B 21 DeepSeek Coder v2 16B 20

Comfortable (10-20 tok/s)

Slow but usable (5-10 tok/s)

Code Llama 34B 9.4 DeepSeek Coder 33B 9.4 Qwen3.8 flash next 125B (A6B) 9.2 DeepSeek-R1 32B 9.0 Qwen2.5 32B 9.0 Qwen 3 32B 9.0 Gemma 4 31B 9.0 Qwen2.5 Coder 32B 9.0 LLaVA 34B 9.0 qwq 32B 9.0 ornith 35B 8.5 Qwen 3-VL 32B 8.5 Qwen2.5-VL 32B 8.5 ornith 1.5 35B 7.8 Nemotron 3 33B 6.4

Too slow (<5 tok/s)

Llama 2 70B 4.6 Code Llama 70B 4.6 orca mini 70B 4.6 Llama 2 uncensored 70B 4.6 Llama 3 70B 4.5 Qwen 72B 4.4 Qwen2 72B 4.4 Llama 3.1 70B 4.2 DeepSeek-R1 70B 4.2 Llama 3.3 70B 4.2 Qwen2.5 72B 3.8 Qwen2.5-VL 72B 3.7 Llama 3.2 Vision 90B 3.3 Qwen 110B 2.8 Mistral medium 3.5 128B 2.2 mixtral 8X22B 2.2