What can I run? · 8K context

Ryzen AI Max+ 395 mini PC (128 GB)

Change hardwareCurrently: Ryzen AI Max+ 395 mini PC (128 GB)

Not sure what you have? Help me checkTwo numbers decide it: memory, and whether there is a graphics card

On a Mac

Click the Apple logo in the top-left corner of the screen, then About This Mac. Note the Chip (M1, M2, M3, M4…) and the Memory (8 GB, 16 GB…).

On Windows

Press Ctrl + Shift + Esc to open Task Manager, then Performance. Memory shows your RAM. If a GPU entry lists Dedicated GPU memory of 4 GB or more, you have a graphics card; otherwise you don’t.

Then pick the closest match — or the Build your own setup option below for anything else:

Mac with 8 GB Mac with 16 GB Windows laptop, 16 GB, no graphics card PC with a 16 GB graphics card

Still unsure? Start with the 16 GB laptop: what runs there runs on almost any recent computer.

Build your own setupChoose GPUs or a chip and how much memory you have

Main deviceHow many

Second device (optional)How many

System RAM (GB)Unified memory (GB, Mac/APU)

18run well

0with offload

1too large

Usable memory~96 GB estimate

Speeds marked measured come from reference results and verified community runs on this exact system; others are estimates.

Start with this one

Qwen3 30B-A3B

Efficient general-purpose model for reasoning, coding and tool use. From Qwen Team (Alibaba Cloud).

Speed: Much faster than you can read51 tok/s measured
Download: 17.4 GiBQ4_K_M
Memory: 19.3 of 96.0 GB estimatefits entirely
License: PermissiveApache License 2.0

Why this one: a general chat model that fits this machine without spilling out of memory, answers faster than you read, and has the strongest benchmark results among those that do.

Now run it

Easiest: LM Studio free desktop app, no terminal

Download LM Studio from lmstudio.ai, install it and open it.
Open the search tab (the magnifying glass) and search for unsloth/Qwen3-30B-A3B-GGUF.
Choose the Q4_K_M download (17.4 GiB) and click Download.
Go to the chat tab, pick Qwen3 30B-A3B at the top, and type your first message.

In a terminal: Ollama one command

Install Ollama from ollama.com/download.
Open a terminal and run:
```
ollama run hf.co/unsloth/Qwen3-30B-A3B-GGUF:Q4_K_M
```
The first run downloads the model (17.4 GiB); then type a message and press Enter.

Everything that runs here grouped by how it runs

Runs in accelerator memory 18

Fits entirely in GPU or unified memory — the fast path.

Model

Recommended download

Memory

Generation speed

Llama 3.1 70B InstructMeta · 70.6B

Q4_K_M via llama.cpp40.2 GiB · LM Studio Community

44.8 of 96.0 GiB estimate

~4 tok/s estimate

Llama 3.3 70B InstructMeta · 70.6B

2 other options

Q3_K_M · Unsloth AI	llama.cpp	Runs well	36.4 GiB	~5 tok/s estimate
AWQ 4-bit · Unsloth AI	SGLang	Runs well	42.3 GiB	~4 tok/s estimate

Q4_K_M via llama.cpp40.2 GiB · LM Studio Community

44.8 of 96.0 GiB estimate

~4 tok/s estimate

Qwen2.5 32B InstructQwen Team (Alibaba Cloud) · 32.8B

3 other options

AWQ 4-bit · Qwen Team (Alibaba Cloud)	SGLang	Runs well	20.8 GiB	~9 tok/s estimate
Q4_K_M · Qwen Team (Alibaba Cloud)	llama.cpp	Runs well	21.9 GiB	~8 tok/s estimate
Q5_K_M · LM Studio Community	llama.cpp	Runs well	25.1 GiB	~7 tok/s estimate

BF16 via SGLang61.0 GiB · Qwen Team (Alibaba Cloud)

66.0 of 96.0 GiB estimate

~3 tok/s estimate

Qwen2.5-Coder 32B InstructQwen Team (Alibaba Cloud) · 32.8B

2 other options

Q4_K_M · Qwen Team (Alibaba Cloud)	llama.cpp	Runs well	21.9 GiB	~8 tok/s estimate
BF16 · Qwen Team (Alibaba Cloud)	SGLang	Runs well	66.0 GiB	~3 tok/s estimate

Q8_0 via llama.cpp32.4 GiB · LM Studio Community

36.2 of 96.0 GiB estimate

~5 tok/s estimate

DeepSeek-R1-Distill-Qwen-32BQwen Team (Alibaba Cloud) · 32.8B

1 other option

Q4_K_M · LM Studio Community	llama.cpp	Runs well	21.9 GiB	~8 tok/s estimate

BF16 via SGLang61.0 GiB · DeepSeek

66.0 of 96.0 GiB estimate

~3 tok/s estimate

Qwen3 30B-A3BQwen Team (Alibaba Cloud) · 30.5B (3.3B active)

5 other options

Q4_K_M · Qwen Team (Alibaba Cloud)	llama.cpp	Runs well	19.2 GiB	~65 tok/s estimate
FP8 · Qwen Team (Alibaba Cloud)	SGLang	Runs well	31.2 GiB	~43 tok/s estimate
Q8_0 · LM Studio Community	llama.cpp	Runs well	32.7 GiB	~41 tok/s estimate
Q8_0 · Qwen Team (Alibaba Cloud)	llama.cpp	Runs well	32.7 GiB	~41 tok/s estimate
BF16 · Qwen Team (Alibaba Cloud)	SGLang	Runs well	60.4 GiB	~23 tok/s estimate

Q4_K_M via llama.cpp17.4 GiB · Unsloth AI

19.3 of 96.0 GiB estimate

51 tok/s measured

Gemma 3 27B ITGoogle DeepMind · 27.4B

1 other option

Q4_K_M · LM Studio Community	llama.cpp	Runs well	20.6 GiB	~10 tok/s estimate

BF16 via SGLang51.1 GiB · Google DeepMind

57.5 of 96.0 GiB estimate

~3 tok/s estimate

Gemma 2 27B ITGoogle DeepMind · 27.2B

1 other option

Q4_K_M · LM Studio Community	llama.cpp	Runs well	19.5 GiB	~10 tok/s estimate

BF16 via SGLang50.7 GiB · Google DeepMind

56.1 of 96.0 GiB estimate

~3 tok/s estimate

Mistral Small 24B Instruct 2501Mistral AI · 23.6B

2 other options

Q4_K_M · LM Studio Community	llama.cpp	Runs well	15.7 GiB	~11 tok/s estimate
Q6_K · LM Studio Community	llama.cpp	Runs well	20.5 GiB	~8 tok/s estimate

BF16 via SGLang43.9 GiB · Mistral AI

47.4 of 96.0 GiB estimate

~4 tok/s estimate

Qwen2.5 14B InstructQwen Team (Alibaba Cloud) · 14.8B

3 other options

AWQ 4-bit · Qwen Team (Alibaba Cloud)	SGLang	Runs well	10.2 GiB	~18 tok/s estimate
Q4_K_M · Qwen Team (Alibaba Cloud)	llama.cpp	Runs well	10.7 GiB	~17 tok/s estimate
Q6_K · LM Studio Community	llama.cpp	Runs well	13.7 GiB	~13 tok/s estimate

BF16 via SGLang27.5 GiB · Qwen Team (Alibaba Cloud)

30.6 of 96.0 GiB estimate

~6 tok/s estimate

Phi-4Microsoft · 14.7B

2 other options

Q4_K_M · LM Studio Community	llama.cpp	Runs well	10.7 GiB	~18 tok/s estimate
BF16 · Microsoft	SGLang	Runs well	30.5 GiB	~6 tok/s estimate

Q8_0 via llama.cpp14.5 GiB · LM Studio Community

17.1 of 96.0 GiB estimate

~10 tok/s estimate

Gemma 2 9B ITGoogle DeepMind · 9.24B

1 other option

Q4_K_M · LM Studio Community	llama.cpp	Runs well	8.6 GiB	~27 tok/s estimate

BF16 via SGLang17.2 GiB · Google DeepMind

21.0 of 96.0 GiB estimate

~9 tok/s estimate

Hermes 3 Llama 3.1 8BMeta · 8.03B

2 other options

Q4_K_M · NousResearch	llama.cpp	Runs well	6.3 GiB	~31 tok/s estimate
Q6_K · NousResearch	llama.cpp	Runs well	7.9 GiB	~23 tok/s estimate

Q8_0 via llama.cpp8.0 GiB · NousResearch

9.8 of 96.0 GiB estimate

~18 tok/s estimate

Llama 3.1 8B InstructMeta · 8.03B

3 other options

Q4_K_M · LM Studio Community	llama.cpp	Runs well	6.3 GiB	~31 tok/s estimate
FP8 · RedHatAI	SGLang	Runs well	10.3 GiB	~17 tok/s estimate
BF16 · Meta	SGLang	Runs well	17.1 GiB	~10 tok/s estimate

Q8_0 via llama.cpp7.9 GiB · LM Studio Community

9.8 of 96.0 GiB estimate

~18 tok/s estimate

Qwen2.5 7B InstructQwen Team (Alibaba Cloud) · 7.62B

3 other options

AWQ 4-bit · Qwen Team (Alibaba Cloud)	SGLang	Runs well	5.2 GiB	~34 tok/s estimate
Q4_K_M · Qwen Team (Alibaba Cloud)	llama.cpp	Runs well	5.5 GiB	~32 tok/s estimate
BF16 · Qwen Team (Alibaba Cloud)	SGLang	Runs well	15.7 GiB	~11 tok/s estimate

Q8_0 via llama.cpp7.5 GiB · Qwen Team (Alibaba Cloud)

8.8 of 96.0 GiB estimate

~19 tok/s estimate

DeepSeek-R1-Distill-Qwen-7BQwen Team (Alibaba Cloud) · 7.62B

1 other option

Q4_K_M · LM Studio Community	llama.cpp	Runs well	5.2 GiB	~32 tok/s estimate

BF16 via SGLang14.2 GiB · DeepSeek

15.5 of 96.0 GiB estimate

~11 tok/s estimate

Mistral 7B Instruct v0.3Mistral AI · 7.25B

2 other options

Q4_K_M · LM Studio Community	llama.cpp	Runs well	5.8 GiB	~34 tok/s estimate
BF16 · Mistral AI	SGLang	Runs well	15.5 GiB	~11 tok/s estimate

Q8_0 via llama.cpp7.2 GiB · LM Studio Community

9.0 of 96.0 GiB estimate

~20 tok/s estimate

Mixtral 8x7B Instruct v0.1Mistral AI · 46.7B (12.9B active)

1 other option

Q4_K_M · LM Studio Community	llama.cpp	Runs well	29.1 GiB	~20 tok/s estimate

BF16 via SGLang87.0 GiB · Mistral AITight fit

92.0 of 96.0 GiB estimate

~6 tok/s estimate

Too large for this hardware1 model variant

DeepSeek-R1 671B · 3 downloads checked

How this is calculatedMemory, fit and speed estimates

Memory = download size + fp16 KV cache for 8K tokens + runtime overhead. Usable memory is 95% of dedicated VRAM, a device-specific share of unified memory (75% by default), and 80% of system RAM for runtimes that can offload. Runtimes must load the file format and support a backend present on the hardware.

Estimated speed is bounded by memory bandwidth ÷ bytes read per token (active parameters for mixture-of-experts), shown in italics with an estimate label and a ±35% range (hover for it). Measured speeds are medians of reference results and verified public community runs on the same system, download and runtime. For each model we recommend the highest-precision download that fits, preferring not to offload.

On a Mac

On Windows

Laptops

Mini PCs

Desktops

Servers

Qwen3 30B-A3B

Now run it

Easiest: LM Studio free desktop app, no terminal

In a terminal: Ollama one command

Everything that runs here grouped by how it runs

Runs in accelerator memory 18