Which local LLMs can the Ryzen AI Max+ 395 (Strix Halo) run?

Unified memory — by default the GPU can use about 70% of it

Assumes 8k context and an f16 KV cache. Verdicts are shown for each memory size.

Specs

Memory options: 32 GB · 64 GB · 128 GB
Memory bandwidth: 256.0 GB/s
FP16 compute: 29.7 TFLOPS
Launch year: 2025
Price: $1,999 launch MSRP (128 GB)
Also written as: Strix Halo, Ryzen AI Max 395, Framework Desktop, 8060S, Radeon 8060S

Verdicts at a glance

At Q4_K_M (or the closest available quant) with 8k context.

32 GBRuns great27 modelsRuns well9 modelsRuns slowly13 modelsWon't run10 models1 more model runs at a lower quant.
64 GBRuns great31 modelsRuns well9 modelsRuns slowly11 modelsWon't run8 models2 more models run at a lower quant.
128 GBRuns great34 modelsRuns well12 modelsRuns slowly10 modelsWon't run3 models2 more models run at a lower quant.

How to read the verdicts

Runs great: Fully on the GPU at 20 tok/s or more
Runs well: Fully on the GPU at 8–20 tok/s; MoE experts in system RAM at 20 tok/s or more; or at least 90% on the GPU at 8 tok/s or more
Runs slowly: 2–8 tok/s; CPU-only; less than 90% on the GPU; or MoE experts in system RAM below 20 tok/s
Won't run: Does not fit, or under 2 tok/s

Every model on the Ryzen AI Max+ 395 (Strix Halo)

Scroll sideways to see every column.

59 models sorted by verdict and speed, with 128 GB of memory
Model	Verdict	Speed	Quant	Memory	Runs as	Context	Notes
MiniCPM5 1B	Runs great	est. 235.5 tok/s188.4–282.6calibrated estimate ±20%	Q4_K_M	1.5 / 89.6 GB	Unified memory	up to 128k	thinking mode
Llama 3.2 1B	Runs great	est. 194.7 tok/s155.7–233.6calibrated estimate ±20%	Q4_K_M	1.7 / 89.6 GB	Unified memory	up to 128k	—
EXAONE 4.0 1.2B	Runs great	est. 159.8 tok/s127.9–191.8calibrated estimate ±20%	Q4_K_M	1.9 / 89.6 GB	Unified memory	up to 64k	thinking mode
Ling-3.0 Tiny	Runs great	est. 154.3 tok/s108.0–200.5theoretical estimate ±30%	Q4_K_M	5.5 / 89.6 GB	Unified memory	up to 128k	thinking mode
LFM2.5 2.6B	Runs great	est. 116.3 tok/s81.4–151.3theoretical estimate ±30%	Q4_K_M	2.4 / 89.6 GB	Unified memory	up to 128k	thinking mode
HyperCLOVA X SEED 1.5B	Runs great	est. 115.6 tok/s92.5–138.8calibrated estimate ±20%	Q4_K_M	2.4 / 89.6 GB	Unified memory	up to 16k	—
MiniCPM5 2B	Runs great	est. 109.8 tok/s87.8–131.7calibrated estimate ±20%	Q4_K_M	2.5 / 89.6 GB	Unified memory	up to 64k	thinking mode
Gemma 4 E2B	Runs great	est. 109.5 tok/s76.6–142.3theoretical estimate ±30%	Q4_K_M	3.7 / 89.6 GB	Unified memory	up to 128k	thinking mode
Mellum2.1 12B-A2.5B Thinking	Runs great	est. 93.1 tok/s74.5–111.7calibrated estimate ±20%	Q4_K_M	8.8 / 89.6 GB	Unified memory	up to 128k	thinking mode
Spark-X2.5 4B	Runs great	est. 72.3 tok/s57.9–86.8calibrated estimate ±20%	Q4_K_M	3.5 / 89.6 GB	Unified memory	up to 64k	thinking mode
Llama 3.2 3B	Runs great	est. 70.9 tok/s56.7–85.1calibrated estimate ±20%	Q4_K_M	3.5 / 89.6 GB	Unified memory	up to 64k	—
gpt-oss-20b	Runs great	est. 68.6 tok/s54.8–82.3calibrated estimate ±20%	MXFP4	12.9 / 89.6 GB	Unified memory	up to 128k	thinking mode
Qwen3.5 4B	Runs great	est. 67.8 tok/s54.2–81.3calibrated estimate ±20%	Q4_K_M	3.7 / 89.6 GB	Unified memory	up to 64k	thinking mode
MiniCPM-V 4.7 35B-A3B	Runs great	est. 67.5 tok/s54.0–81.1calibrated estimate ±20%	Q4_K_M	21.9 / 89.6 GB	Unified memory	up to 128k	thinking mode
Nemotron 3.5 Lightning 30B-A3B	Runs great	est. 65.9 tok/s46.1–85.7theoretical estimate ±30%	Q4_K_M	26.1 / 89.6 GB	Unified memory	up to 256k	thinking mode
Qwen3.5 35B-A3B	Runs great	64.8 tok/s8k estimate 56.0 tok/s (44.8–67.2, calibrated estimate ±20%)measured (1 run, 512 context)	Q4_K_M	23.4 / 89.6 GB	Unified memory	up to 128k	thinking mode
Qwen3.6 35B-A3B	Runs great	63.8 tok/s8k estimate 63.1 tok/s (50.5–75.7, calibrated estimate ±20%)measured (1 run, 512 context)	Q4_K_M	23.0 / 89.6 GB	Unified memory	up to 128k	thinking mode
Gemma 4 E4B	Runs great	est. 59.3 tok/s41.5–77.1theoretical estimate ±30%	Q4_K_M	5.7 / 89.6 GB	Unified memory	up to 128k	thinking mode
Kanana 1.5 15.7B-A3B	Runs great	est. 57.2 tok/s45.8–68.6calibrated estimate ±20%	Q4_K_M	12.1 / 89.6 GB	Unified memory	up to 32k	—
Gemma 4 26B-A4B	Runs great	54.3 tok/s8k estimate 48.1 tok/s (38.5–57.7, calibrated estimate ±20%)measured (2 runs, 512 context)	Q4_K_M	17.7 / 89.6 GB	Unified memory	up to 128k	thinking mode
Qwen3 30B-A3B (2507)	Runs great	est. 54.2 tok/s43.3–65.0calibrated estimate ±20%	Q4_K_M	19.9 / 89.6 GB	Unified memory	up to 64k	—
gpt-oss-120b	Runs great	50.2 tok/s8k estimate 48.5 tok/s (38.8–58.1, calibrated estimate ±20%)measured (1 run, 512 context)	MXFP4	64.3 / 89.6 GB	Unified memory	up to 64k	thinking mode
Ling-3.0 Flash	Runs great	est. 48.4 tok/s33.9–62.9theoretical estimate ±30%	Q4_K_M	77.7 / 89.6 GB	Unified memory	up to 256k	thinking mode
Kolibri 1	Runs great	est. 48.0 tok/s38.4–57.6calibrated estimate ±20%	Q4_K_M	50.5 / 89.6 GB	Unified memory	up to 128k	thinking mode
Qwen3 8B	Runs great	43.6 tok/s8k estimate 33.7 tok/s (26.9–40.4, calibrated estimate ±20%)measured (1 run, 512 context)	Q4_K_M	6.8 / 89.6 GB	Unified memory	up to 8k	thinking mode
Qwen3.5 9B	Runs great	37.2 tok/s8k estimate 34.2 tok/s (27.4–41.0, calibrated estimate ±20%)measured (1 run, 4k context)	Q4_K_M	6.7 / 89.6 GB	Unified memory	up to 32k	thinking mode
K2-Horizon MoVA 36B-A4B	Runs great	est. 35.1 tok/s24.5–45.6theoretical estimate ±30%	Q4_K_M	24.6 / 89.6 GB	Unified memory	up to 8k	thinking mode
DeepSeek R1 Distill Llama 8B	Runs great	est. 35.0 tok/s28.0–42.0calibrated estimate ±20%	Q4_K_M	6.6 / 89.6 GB	Unified memory	up to 8k	thinking mode
Kanana 1.5 8B	Runs great	est. 35.0 tok/s28.0–42.0calibrated estimate ±20%	Q4_K_M	6.6 / 89.6 GB	Unified memory	up to 32k	—
Llama 3.1 8B	Runs great	est. 35.0 tok/s28.0–42.0calibrated estimate ±20%	Q4_K_M	6.6 / 89.6 GB	Unified memory	up to 32k	—
Nanbeige4.2 3B	Runs great	est. 34.3 tok/s24.0–44.5theoretical estimate ±30%	Q4_K_M	4.7 / 89.6 GB	Unified memory	up to 8k	thinking mode
DeepSeek R1 0528 Qwen3 8B	Runs great	est. 33.7 tok/s26.9–40.4calibrated estimate ±20%	Q4_K_M	6.8 / 89.6 GB	Unified memory	up to 8k	thinking mode
K2-Horizon 7B	Runs great	est. 30.9 tok/s24.7–37.1calibrated estimate ±20%	Q4_K_M	7.4 / 89.6 GB	Unified memory	up to 8k	thinking mode
Gemma 3 12B	Runs great	est. 26.8 tok/s21.4–32.1calibrated estimate ±20%	Q4_K_M	8.4 / 89.6 GB	Unified memory	up to 32k	—
Laguna S 2.1	Runs well	est. 26.4 tok/s21.1–31.7calibrated estimate ±20%	Q4_K_M	68.6 / 89.6 GB	Unified memory	up to 128k	thinking mode
Gemma 4 12B	Runs well	24.4 tok/s8k estimate 28.9 tok/s (23.2–34.7, calibrated estimate ±20%)measured (1 run, 512 context)	Q4_K_M	7.8 / 89.6 GB	Unified memory	up to 256k	UD-Q2_K_XL (heavy quality loss): Runs great thinking mode
Solar Open 100B	Runs well	est. 21.6 tok/s17.3–26.0calibrated estimate ±20%	Q4_K_M	64.4 / 89.6 GB	Unified memory	up to 32k	thinking mode
Qwen3.5 122B-A10B	Runs well	20.9 tok/s8k estimate 22.7 tok/s (18.2–27.2, calibrated estimate ±20%)measured (1 run, 512 context)	Q4_K_M	79.0 / 89.6 GB	Unified memory	up to 256k	UD-Q2_K_XL (heavy quality loss): Runs great thinking mode
HyperCLOVA X SEED Think 14B	Runs well	est. 20.6 tok/s14.4–26.8theoretical estimate ±30%	Q4_K_M	10.8 / 89.6 GB	Unified memory	up to 32k	thinking mode
Qwen3 14B	Runs well	est. 20.3 tok/s16.2–24.4calibrated estimate ±20%	Q4_K_M	10.9 / 89.6 GB	Unified memory	up to 32k	thinking mode
DeepSeek R1 Distill Qwen 14B	Runs well	est. 19.8 tok/s15.8–23.8calibrated estimate ±20%	Q4_K_M	11.2 / 89.6 GB	Unified memory	up to 32k	thinking mode
Phi-4	Runs well	est. 19.6 tok/s15.7–23.5calibrated estimate ±20%	Q4_K_M	11.3 / 89.6 GB	Unified memory	up to 16k	—
Mistral Small 3.2 24B	Runs well	est. 13.4 tok/s10.7–16.1calibrated estimate ±20%	Q4_K_M	16.2 / 89.6 GB	Unified memory	up to 64k	Q2_K (heavy quality loss): Runs great
Gemma 3 27B	Runs well	est. 12.2 tok/s9.8–14.6calibrated estimate ±20%	Q4_K_M	17.8 / 89.6 GB	Unified memory	up to 64k	—
Qwen3.6 27B	Runs well	est. 12.1 tok/s9.7–14.5calibrated estimate ±20%	Q4_K_M	17.9 / 89.6 GB	Unified memory	up to 8k	thinking mode
Muse Glimmer 30B	Runs well	est. 12.1 tok/s9.6–14.5calibrated estimate ±20%	Q4_K_M	18.0 / 89.6 GB	Unified memory	up to 8k	thinking mode
Qwen3.5 27B	Runs slowly	est. 11.9 tok/s9.5–14.3calibrated estimate ±20%	Q4_K_M	18.2 / 89.6 GB	Unified memory	up to 256k	Try IQ4_XS: Runs well thinking mode
Qwen3.8 27B	Runs slowly	est. 11.7 tok/s9.3–14.0calibrated estimate ±20%	Q4_K_M	18.6 / 89.6 GB	Unified memory	up to 256k	Try IQ4_XS: Runs well thinking mode
Gemma 4 31B	Runs slowly	est. 11.1 tok/s8.8–13.3calibrated estimate ±20%	Q4_K_M	19.6 / 89.6 GB	Unified memory	up to 256k	Try IQ4_XS: Runs well thinking mode
Granite 4.2 30B	Runs slowly	est. 10.6 tok/s8.5–12.7calibrated estimate ±20%	Q4_K_M	20.4 / 89.6 GB	Unified memory	up to 128k	Q2_K (heavy quality loss): Runs well thinking mode
EXAONE 4.0 32B	Runs slowly	est. 10.6 tok/s8.4–12.7calibrated estimate ±20%	Q4_K_M	20.5 / 89.6 GB	Unified memory	up to 128k	thinking mode
EXAONE 4.5 33B	Runs slowly	est. 10.2 tok/s8.2–12.2calibrated estimate ±20%	Q4_K_M	21.2 / 89.6 GB	Unified memory	up to 256k	thinking mode
Qwen3 32B	Runs slowly	10.2 tok/s8k estimate 9.6 tok/s (7.7–11.5, calibrated estimate ±20%)measured (2 runs, 512 context)	Q4_K_M	22.5 / 89.6 GB	Unified memory	up to 32k	Q2_K (heavy quality loss): Runs well thinking mode
DeepSeek R1 Distill Qwen 32B	Runs slowly	est. 9.5 tok/s7.6–11.5calibrated estimate ±20%	Q4_K_M	22.6 / 89.6 GB	Unified memory	up to 128k	Q2_K (heavy quality loss): Runs well thinking mode
Nanbeige4.3 33B	Runs slowly	est. 9.3 tok/s7.5–11.2calibrated estimate ±20%	Q4_K_M	23.1 / 89.6 GB	Unified memory	up to 128k	Q2_K (heavy quality loss): Runs well thinking mode
Llama 3.3 70B	Runs slowly	est. 4.6 tok/s3.7–5.6calibrated estimate ±20%	Q4_K_M	45.8 / 89.6 GB	Unified memory	up to 128k	—
DeepSeek V4 Flash 0731	Won't runTry Q2_K (heavy quality loss): Runs slowly	est. 19.1 tok/s13.4–24.9theoretical estimate ±30%	Q2_K	117.4 GB RAM	CPU only	up to 256k	CPU inference — capped at Runs slowly thinking mode
Solar Open 2 250B	Won't runTry Q2_K (heavy quality loss): Runs slowly	est. 18.0 tok/s12.6–23.4theoretical estimate ±30%	Q2_K	95.8 GB RAM	CPU only	up to 256k	CPU inference — capped at Runs slowly thinking mode
GLM-5.3 Flash	Won't run	—	Q4_K_M	needs 182.1 GB	—	—	Needs about 182.1 GB; 124.0 GB of memory is free (89.6 GB usable by the GPU) thinking mode

Models with a thinking mode spend extra tokens thinking, so their speed thresholds are 1.5× stricter (30 / 12 / 3 tok/s). With thinking turned off, a model may do one grade better.

Verdict by memory size

Q4_K_M verdict for each memory configuration — open the model page for speeds
Model	32 GB	64 GB	128 GB
MiniCPM5 1B	Runs great	Runs great	Runs great
Llama 3.2 1B	Runs great	Runs great	Runs great
EXAONE 4.0 1.2B	Runs great	Runs great	Runs great
Ling-3.0 Tiny	Runs great	Runs great	Runs great
LFM2.5 2.6B	Runs great	Runs great	Runs great
HyperCLOVA X SEED 1.5B	Runs great	Runs great	Runs great
MiniCPM5 2B	Runs great	Runs great	Runs great
Gemma 4 E2B	Runs great	Runs great	Runs great
Mellum2.1 12B-A2.5B Thinking	Runs great	Runs great	Runs great
Spark-X2.5 4B	Runs great	Runs great	Runs great
Llama 3.2 3B	Runs great	Runs great	Runs great
gpt-oss-20b	Runs great	Runs great	Runs great
Qwen3.5 4B	Runs great	Runs great	Runs great
MiniCPM-V 4.7 35B-A3B	Runs great	Runs great	Runs great
Nemotron 3.5 Lightning 30B-A3B	Runs slowly	Runs great	Runs great
Qwen3.5 35B-A3B	Runs slowly	Runs great	Runs great
Qwen3.6 35B-A3B	Runs slowly	Runs great	Runs great
Gemma 4 E4B	Runs great	Runs great	Runs great
Kanana 1.5 15.7B-A3B	Runs great	Runs great	Runs great
Gemma 4 26B-A4B	Runs great	Runs great	Runs great
Qwen3 30B-A3B (2507)	Runs great	Runs great	Runs great
gpt-oss-120b	Won't run	Won't run	Runs great
Ling-3.0 Flash	Won't run	Won't run	Runs great
Kolibri 1	Won't run	Runs slowly	Runs great
Qwen3 8B	Runs great	Runs great	Runs great
Qwen3.5 9B	Runs great	Runs great	Runs great
K2-Horizon MoVA 36B-A4B	Runs slowly	Runs great	Runs great
DeepSeek R1 Distill Llama 8B	Runs great	Runs great	Runs great
Kanana 1.5 8B	Runs great	Runs great	Runs great
Llama 3.1 8B	Runs great	Runs great	Runs great
Nanbeige4.2 3B	Runs great	Runs great	Runs great
DeepSeek R1 0528 Qwen3 8B	Runs great	Runs great	Runs great
K2-Horizon 7B	Runs great	Runs great	Runs great
Gemma 3 12B	Runs great	Runs great	Runs great
Laguna S 2.1	Won't run	Won't run	Runs well
Gemma 4 12B	Runs well	Runs well	Runs well
Solar Open 100B	Won't run	Won't run	Runs well
Qwen3.5 122B-A10B	Won't run	Won't run	Runs well
HyperCLOVA X SEED Think 14B	Runs well	Runs well	Runs well
Qwen3 14B	Runs well	Runs well	Runs well
DeepSeek R1 Distill Qwen 14B	Runs well	Runs well	Runs well
Phi-4	Runs well	Runs well	Runs well
Mistral Small 3.2 24B	Runs well	Runs well	Runs well
Gemma 3 27B	Runs well	Runs well	Runs well
Qwen3.6 27B	Runs well	Runs well	Runs well
Muse Glimmer 30B	Runs well	Runs well	Runs well
Qwen3.5 27B	Runs slowly	Runs slowly	Runs slowly
Qwen3.8 27B	Runs slowly	Runs slowly	Runs slowly
Gemma 4 31B	Runs slowly	Runs slowly	Runs slowly
Granite 4.2 30B	Runs slowly	Runs slowly	Runs slowly
EXAONE 4.0 32B	Runs slowly	Runs slowly	Runs slowly
EXAONE 4.5 33B	Runs slowly	Runs slowly	Runs slowly
Qwen3 32B	Runs slowly	Runs slowly	Runs slowly
DeepSeek R1 Distill Qwen 32B	Runs slowly	Runs slowly	Runs slowly
Nanbeige4.3 33B	Runs slowly	Runs slowly	Runs slowly
Llama 3.3 70B	Won't run	Runs slowly	Runs slowly
DeepSeek V4 Flash 0731	Won't run	Won't run	Won't run
Solar Open 2 250B	Won't run	Won't run	Won't run
GLM-5.3 Flash	Won't run	Won't run	Won't run

Measured results on the Ryzen AI Max+ 395 (Strix Halo)

Public benchmarks we calibrate against. Their conditions (context, backend, flags) can differ from the estimates above.

Measured results on the Ryzen AI Max+ 395 (Strix Halo)
Model	Quant	Backend	Context	Prompt (tok/s)	Generation (tok/s)	Flags	Source	Measured
gpt-oss-120b	MXFP4	llama.cpp	512	707.0	50.2	ROCm 7.0.2, no mmap	quozul.dev	2025-10-27
Gemma 4 26B-A4B	Q4_K_M	llama.cpp	512	1,327.0	55.5	Vulkan	github.com	2026-06-30
Gemma 4 26B-A4B	Q4_K_M	llama.cpp	512	1,209.0	53.0	Vulkan	github.com	2026-08-18
Gemma 4 26B-A4B	UD-Q4_K_XL	llama.cpp	512	1,325.0	54.7	Vulkan	github.com	2026-05-18
Gemma 4 31B	UD-Q4_K_XL	llama.cpp	512	302.0	11.5	Vulkan	github.com	2026-05-18
Gemma 4 12B	IQ4_XS	llama.cpp	512	680.0	25.7	Vulkan	github.com	2026-06-04
Gemma 4 12B	Q4_K_M	llama.cpp	512	685.0	24.4	Vulkan	github.com	2026-06-04
Qwen3.6 35B-A3B	Q8_0	llama.cpp	512	—	50.6	Vulkan, FA off	github.com	2026-06-10
Qwen3.6 35B-A3B	UD-Q8_K_XL	llama.cpp	512	1,046.0	46.2	Vulkan	github.com	2026-04-26
Qwen3.6 35B-A3B	Q4_K_M	llama.cpp	512	1,064.0	63.8	Vulkan	github.com	2026-04-26
Qwen3.5 35B-A3B	Q4_K_M	llama.cpp	512	1,077.0	64.8	Vulkan	github.com	2026-04-26
Qwen3.5 9B	Q4_K_M	llama.cpp	4k	969.0	37.2	Vulkan	localmaxxing.com	2026-04-26
Qwen3.5 122B-A10B	Q4_K_M	llama.cpp	512	315.0	20.9	KV q8_0	github.com	2026-07-19
Qwen3 8B	Q4_K_M	llama.cpp	512	1,238.0	43.6	Vulkan	github.com	2026-09-08
Qwen3 32B	Q4_K_M	llama.cpp	512	256.0	9.9	ROCm	blog.tosukui.xyz	2025-07-13
Qwen3 32B	Q4_K_M	llama.cpp	512	140.0	10.4	Vulkan	blog.tosukui.xyz	2025-05-16
Qwen3 32B	Q8_0	llama.cpp	512	77.0	6.4	ROCm	llm-tracker.info	2025-05-11

Frequently asked questions

What is the largest model that runs entirely on the Ryzen AI Max+ 395 (Strix Halo)?

Ling-3.0 Flash at Q4_K_M (a 77.01 GB file) fits entirely in 128 GB of unified memory with 8k context, at est. 48.4 tok/s (33.9–62.9, theoretical estimate ±30%).

How many local LLMs run well on the Ryzen AI Max+ 395 (Strix Halo)?

At Q4_K_M with 8k context, with 128 GB of memory, out of 59 tracked models: 34 run great, 12 run well, 10 run slowly and 3 won't run.

Can the Ryzen AI Max+ 395 (Strix Halo) run a 70B model like Llama 3.3 70B?

Runs slowly — Unified memory, Q4_K_M: est. 4.6 tok/s (3.7–5.6, calibrated estimate ±20%).

Speeds are estimates from memory bandwidth, calibrated against public benchmarks, and each one comes with an error band and a confidence label. Real results vary with drivers, backend, context length and thermals.