Menu
mutinaiPreview: Public preview · figures are illustrative fixture data, attributed to their source · data to May 2025 · read-only
What can I run?
What can I run? · 8K context

RTX 4090 workstation (64 GB DDR5)

Change hardwareCurrently: RTX 4090 workstation (64 GB DDR5)
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 GBMac with 16 GBWindows laptop, 16 GB, no graphics cardPC with a 16 GB graphics card

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

Laptops

MacBook Air M1 8 GBMacBook Air M2 16 GBMacBook Air M3 16 GBMacBook Air M4 16 GBMacBook Pro M3 Max 64 GBMacBook Pro M4 Max 128 GBWindows laptop 16 GB

Mini PCs

GB10 mini workstationMac mini M4 16 GBMac mini M4 Pro 48 GBRyzen AI Max+ 395 mini PC

Desktops

CPU-only Ryzen 9 7950XDual RTX 3090Mac Studio M2 Ultra 192 GBRTX 4060 Ti 16GB budget buildRTX 4090 workstationRTX 5090 workstationRX 7900 XTX desktop

Servers

A100 80GB server
Build your own setupChoose GPUs or a chip and how much memory you have
15run well
3with offload
1too large

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 read152 tok/s measured
Download
17.4 GiBQ4_K_M
Memory
19.3 of 22.8 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

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

In a terminal: Ollama one command

  1. Install Ollama from ollama.com/download.
  2. 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 15

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

ModelRecommended downloadMemoryGeneration speed
Variant Qwen3 30B-A3BQwen Team (Alibaba Cloud) · 30.5B (3.3B active)
3 other options
Q4_K_M · Qwen Team (Alibaba Cloud)llama.cppRuns well19.2 GiB~258 tok/s estimate
Q8_0 · LM Studio Communityllama.cppWith offload32.7 GiB~40 tok/s estimate
Q8_0 · Qwen Team (Alibaba Cloud)llama.cppWith offload32.7 GiB~39 tok/s estimate
Q4_K_M via llama.cpp17.4 GiB · Unsloth AI
19.3 of 22.8 GiB estimate
152 tok/s measured
Variant Gemma 2 27B ITGoogle DeepMind · 27.2BQ4_K_M via llama.cpp15.5 GiB · LM Studio Community
19.5 of 22.8 GiB estimate
~38 tok/s estimate
Variant Mistral Small 24B Instruct 2501Mistral AI · 23.6B
1 other option
Q4_K_M · LM Studio Communityllama.cppRuns well15.7 GiB~44 tok/s estimate
Q6_K via llama.cpp18.0 GiB · LM Studio Community
20.5 of 22.8 GiB estimate
~33 tok/s estimate
Variant Qwen2.5 14B InstructQwen Team (Alibaba Cloud) · 14.8B
2 other options
AWQ 4-bit · Qwen Team (Alibaba Cloud)SGLangRuns well10.2 GiB~73 tok/s estimate
Q4_K_M · Qwen Team (Alibaba Cloud)llama.cppRuns well10.7 GiB~69 tok/s estimate
Q6_K via llama.cpp11.3 GiB · LM Studio Community
13.7 of 22.8 GiB estimate
~52 tok/s estimate
Variant Phi-4Microsoft · 14.7B
1 other option
Q4_K_M · LM Studio Communityllama.cppRuns well10.7 GiB~69 tok/s estimate
Q8_0 via llama.cpp14.5 GiB · LM Studio Community
17.1 of 22.8 GiB estimate
~41 tok/s estimate
Variant Hermes 3 Llama 3.1 8BMeta · 8.03B
2 other options
Q4_K_M · NousResearchllama.cppRuns well6.3 GiB~120 tok/s estimate
Q6_K · NousResearchllama.cppRuns well7.9 GiB~92 tok/s estimate
Q8_0 via llama.cpp8.0 GiB · NousResearch
9.8 of 22.8 GiB estimate
~72 tok/s estimate
Variant Llama 3.1 8B InstructMeta · 8.03B
3 other options
Q8_0 · LM Studio Communityllama.cppRuns well9.8 GiB~72 tok/s estimate
FP8 · RedHatAISGLangRuns well10.3 GiB~68 tok/s estimate
BF16 · MetaSGLangRuns well17.1 GiB~40 tok/s estimate
Q4_K_M via llama.cpp4.6 GiB · LM Studio Community
6.3 of 22.8 GiB estimate
128 tok/s measured
Variant Qwen2.5 7B InstructQwen Team (Alibaba Cloud) · 7.62B
3 other options
AWQ 4-bit · Qwen Team (Alibaba Cloud)SGLangRuns well5.2 GiB~133 tok/s estimate
Q4_K_M · Qwen Team (Alibaba Cloud)llama.cppRuns well5.5 GiB~126 tok/s estimate
BF16 · Qwen Team (Alibaba Cloud)SGLangRuns well15.7 GiB~42 tok/s estimate
Q8_0 via llama.cpp7.5 GiB · Qwen Team (Alibaba Cloud)
8.8 of 22.8 GiB estimate
~76 tok/s estimate
Variant DeepSeek-R1-Distill-Qwen-7BQwen Team (Alibaba Cloud) · 7.62B
1 other option
Q4_K_M · LM Studio Communityllama.cppRuns well5.2 GiB~126 tok/s estimate
BF16 via SGLang14.2 GiB · DeepSeek
15.5 of 22.8 GiB estimate
~42 tok/s estimate
Variant Mistral 7B Instruct v0.3Mistral AI · 7.25B
2 other options
Q4_K_M · LM Studio Communityllama.cppRuns well5.8 GiB~132 tok/s estimate
BF16 · Mistral AISGLangRuns well15.5 GiB~44 tok/s estimate
Q8_0 via llama.cpp7.2 GiB · LM Studio Community
9.0 of 22.8 GiB estimate
~80 tok/s estimate
Variant Qwen2.5 32B InstructQwen Team (Alibaba Cloud) · 32.8B
2 other options
AWQ 4-bit · Qwen Team (Alibaba Cloud)SGLangTight fit20.8 GiB~34 tok/s estimate
Q5_K_M · LM Studio Communityllama.cppWith offload25.1 GiB~14 tok/s estimate
Q4_K_M via llama.cpp18.7 GiB · Qwen Team (Alibaba Cloud)Tight fit
21.9 of 22.8 GiB estimate
~32 tok/s estimate
Variant Qwen2.5-Coder 32B InstructQwen Team (Alibaba Cloud) · 32.8B
1 other option
Q8_0 · LM Studio Communityllama.cppWith offload36.2 GiB~4 tok/s estimate
Q4_K_M via llama.cpp18.7 GiB · Qwen Team (Alibaba Cloud)Tight fit
21.9 of 22.8 GiB estimate
~32 tok/s estimate
Variant DeepSeek-R1-Distill-Qwen-32BQwen Team (Alibaba Cloud) · 32.8BQ4_K_M via llama.cpp18.7 GiB · LM Studio CommunityTight fit
21.9 of 22.8 GiB estimate
~32 tok/s estimate
Variant Gemma 3 27B ITGoogle DeepMind · 27.4BQ4_K_M via llama.cpp15.6 GiB · LM Studio CommunityTight fit
20.6 of 22.8 GiB estimate
~38 tok/s estimate
Variant Gemma 2 9B ITGoogle DeepMind · 9.24B
1 other option
Q4_K_M · LM Studio Communityllama.cppRuns well8.6 GiB~106 tok/s estimate
BF16 via SGLang17.2 GiB · Google DeepMindTight fit
21.0 of 22.8 GiB estimate
~34 tok/s estimate

Runs with partial offload to system RAM 3

Too big for accelerator memory alone; part of it runs from system RAM, which is much slower.

ModelRecommended downloadMemoryGeneration speed
Variant Llama 3.1 70B InstructMeta · 70.6BQ4_K_M via llama.cpp40.2 GiB · LM Studio Community
44.8 of 22.8 GiB · 22.0 in RAM estimate
~2 tok/s estimate
Variant Llama 3.3 70B InstructMeta · 70.6B
1 other option
Q4_K_M · LM Studio Communityllama.cppWith offload44.8 GiB~2 tok/s estimate
Q3_K_M via llama.cpp32.1 GiB · Unsloth AI
36.4 of 22.8 GiB · 13.6 in RAM estimate
~4 tok/s estimate
Variant Mixtral 8x7B Instruct v0.1Mistral AI · 46.7B (12.9B active)Q4_K_M via llama.cpp26.6 GiB · LM Studio Community
29.1 of 22.8 GiB · 6.3 in RAM estimate
~24 tok/s estimate
Too large for this hardware1 model variant
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.