Hardware

AMD Ryzen AI Max+ 395

AMD Ryzen AI Max+ 395 has 32 / 64 / 96 / 128 GB LPDDR5X and 256 GB/s. See which open models fit, at which quant, and roughly how fast. Figures below use the default memory (64 GB). Updated 2026-10-24.

Memory32 / 64 / 96 / 128 GB LPDDR5X

Bandwidth256 GB/s

Kindunified

Year2025

Strix Halo class APU with a large unified pool. 256 GB/s assumes 256-bit LPDDR5X-8000. The 75% GPU budget is a conservative default, not an AMD firmware limit. Source.

RSS for new models on this device

Models that fit

Model	Verdict	Quant	Context	tok/s
Qwen3 Next 80B A3B Instruct 79.7B	Tight fit	UD-Q4_K_XL	8K	55–76
Qwen3 Next 80B A3B Thinking 79.7B	Tight fit	UD-Q4_K_XL	8K	56–76
Qwen3 Coder Next 79.7B	Tight fit	Q4_K_M	8K	16–21
Mistral Small 4 119B 2603 119B	Tight fit	UD-IQ3_S	8K	17–23
Qwen3.5 122B A10B MTP 125B	Tight fit	UD-IQ3_XXS	8K	35–48
Llama 4 Scout 17B 16E Instruct 108B	Tight fit	Q3_K_S	8K	12–17
GLM 4.5 Air 110B	Tight fit	UD-Q2_K_XL	8K	12–17
GLM 4.6V 107B	Tight fit	UD-Q2_K_XL	8K	12–17
Kwaipilot KAT Coder V2.5 Dev 34.7B	Runs	Q8_0	8K	21–28
Qwen3.8 35B A3B Distill 35.5B	Runs	Q8_0	8K	42–57
Cyber Tiel Coder 35B A3B GGUF MTP 35.5B	Runs	Q8_0 (est.)	8K	42–57
Qwen3.6 35B A3B 34.7B	Runs	UD-Q8_K_XL	8K	40–55
Tiel Coder 35B A3B 34.7B	Runs	UD-Q8_K_XL	8K	40–55
Qwen3.5 35B A3B 34.7B	Tight fit	UD-Q8_K_XL	8K	32–44
NVIDIA Nemotron 3.5 Lightning 30B A3B 32.9B	Runs	UD-Q8_K_XL	8K	36–49
NVIDIA Nemotron 3 Nano Omni 30B A3B Reasoning 31.6B	Runs	UD-Q8_K_XL	8K	37–51
Nemotron 3 Nano 30B A3B 31.6B	Runs	UD-Q8_K_XL	8K	33–45
GLM 4.7 Flash 29.9B	Runs	UD-Q8_K_XL	8K	18–24
Qwen3 Coder 30B A3B Instruct 30.5B	Runs	UD-Q8_K_XL	8K	32–44
Qwen3 30B A3B Instruct 2507 30.5B	Runs	UD-Q8_K_XL	8K	32–44
Qwen3 VL 30B A3B Instruct 30.5B	Runs	UD-Q8_K_XL	8K	32–44
Qwen3 30B A3B Thinking 2507 30.5B	Runs	UD-Q8_K_XL	8K	32–44
bottlecapai ThinkingCap gemma 4 26B A4B it 25.2B	Runs great	Q8_0	8K	22–31
gemma 4 26B A4B it 25.2B	Runs great	UD-Q8_K_XL	8K	22–30
LFM2.5 24B A2B 24.0B	Tight fit	BF16	8K	30–41
GLM 4.7 Flash REAP 23B A3B 23.0B	Tight fit	BF16	8K	22–30
gpt oss 20b 20.9B	Runs great	UD-Q8_K_XL	8K	35–48
Qwen3.6 14B A3B FableVibes 13.8B	Runs great	Q8_0	8K	42–57
gemma 4 E4B it qat 7.5B	Runs great	UD-Q4_K_XL	8K	29–39
LFM2.5 8B A1B 8.5B	Runs great	BF16	8K	59–80
gemma 4 E2B it 4.6B	Runs great	BF16	8K	15–20
Qwen3.8 4B Distill 4.3B	Runs great	BF16	8K	16–21
Qwen3.5 4B 4.2B	Runs great	BF16	8K	16–22
Sharp Spark X2.5 4B 4.1B	Runs great	Q6_K	8K	29–40
Qwen3 4B 4.0B	Runs great	BF16	8K	15–21
Qwen3 VL 4B Instruct 4.0B	Runs great	BF16	8K	15–21
Qwen3 4B Thinking 2507 4.0B	Runs great	F16	8K	15–21
Qwen3 VL 4B Thinking 4.0B	Runs great	BF16	8K	15–21
NVIDIA Nemotron 3 Nano 4B 4.0B	Runs great	BF16	8K	15–20
gemma 3 4b it 3.9B	Runs great	BF16	8K	16–22

105 more models in the catalog do not fit this default profile, or ranked below the first 40 that do.

Example: Tight fit fully in unified memory on your AMD Ryzen AI Max+ 395 at UD-Q4_K_XL, 8K context, about 55–76 tok/s. Weights 42.9 GB plus 0.75 GB of KV cache — 44.4 GB of 48.0 GB usable. Up to 32K context still fits. Weights shown in the verdict use GiB (the same 1024³ counting as GPU memory).