Skip to content

What can the Mac mini M4 24 GB run?

What fits on the Mac mini M4 24 GB
Text models: 246Visual models: 12Strongest first
Memory
24 GB unified
Bandwidth
120 GB/s
Maker
Apple

Open the calculator on this machine

Text models that run on it

Text models that run on the Mac mini M4 24 GB, best first
ModelBest fitContextSpeedStrongest at
Qwen3.8 27B3-bit24K7.2 tok/sChat, Coding, Agents, Docs
Muse Glimmer 30B3-bit56K6.9 tok/sDocs
Gemma 4 26B A4B3-bit240K43 tok/sChat, Coding, Agents, Docs
Qwen3.5 9B8-bit152K10 tok/sChat, Coding, Agents, Docs
Gemma 4 12B8-bit256K7.5 tok/sChat, Coding, Docs
Nemotron 3.5 Lightning 30B A3B3-bit32K44 tok/sAgents, Docs
Qwen3.5 4B8-bit256K19 tok/sAgents, Docs
Devstral Small 2 24B3-bit20K8.4 tok/sCoding
LFM2.5 24B A2B4-bit56K56 tok/sAgents
Mistral Small 3.2 24B3-bit20K8.4 tok/sChat, Coding, Agents
Ministral 3 14B6-bit24K8.8 tok/sCoding, Agents, Docs
Gemma 4 E4B8-bit128K17 tok/sChat, Coding, Docs
gpt-oss 20BMXFP4128K34 tok/sAgents
Ministral 3 8B8-bit40K11 tok/sDocs
Gemma 4 E2B8-bit128K31 tok/sChat
Nemotron 3 Nano 4B8-bit256K23 tok/sDocs
Llama 3.1 8B8-bit52K12 tok/sChat, Coding, Agents, Docs
Qwen3 0.6B8-bit40K97 tok/sNot rated
Qwen3 14B6-bit20K8.3 tok/sNot rated
Qwen3 VL 8B Instruct8-bit40K11 tok/sText and images
Qwen3 8B8-bit40K12 tok/sNot rated
Qwen2.5 7B Instruct8-bit32K13 tok/sNot rated
Qwen2.5 VL 7B Instruct8-bit112K12 tok/sText and images
Unlimited OCR8-bit32K71 tok/sReading documents
Qwen3 VL 4B Instruct8-bit72K20 tok/sText and images
Qwen3 1.7B8-bit40K41 tok/sNot rated
Qwen3 4B8-bit40K22 tok/sNot rated
Qwen3 4B Instruct 25078-bit72K22 tok/sNot rated
Qwen2.5 3B Instruct8-bit32K28 tok/sNot rated
Qwen3.6 27B3-bit24K7.2 tok/sText and images
DeepSeek OCR8-bit8K71 tok/sReading documents
Qwen2.5 Coder 14B Instruct6-bit16K8.3 tok/sCoding
Qwen2.5 VL 3B Instruct8-bit125K23 tok/sText and images
Qwen3.5 27B3-bit24K7.2 tok/sText and images
NVIDIA Nemotron 3 Nano 30B A3B BF163-bit32K44 tok/sNot rated
Qwen2.5 Coder 7B Instruct8-bit32K13 tok/sCoding
Mistral 7B Instruct V0.28-bit32K13 tok/sNot rated
Chandra OCR 28-bit256K17 tok/sReading documents
Qwen3 VL 2B Instruct8-bit112K40 tok/sText and images
MiniCPM5 2B8-bit128K36 tok/sNot rated
Gemma 3 4B IT8-bit128K24 tok/sText and images
Llama 3.2 3B Instruct8-bit104K27 tok/sNot rated
Qwen2.5 14B Instruct6-bit16K8.3 tok/sNot rated
Qwen3 1.7B Base8-bit32K48 tok/sNot rated
Diffusiongemma 26B A4B IT3-bit120K42 tok/sText and images
HunyuanOCR8-bit128K71 tok/sText and images
Qwen3 0.6B Base8-bit32K115 tok/sNot rated
Qwen2 VL 7B Instruct8-bit32K12 tok/sText and images
Ornith 1.5 9B8-bit144K10 tok/sNot rated
Unlimited OCR 28-bit128K71 tok/sReading documents
Granite 4.1 3B8-bit128K25 tok/sNot rated
Mistral 7B V0.18-bit32K13 tok/sNot rated
Qwen2.5 Coder 3B Instruct8-bit32K28 tok/sCoding
Meta Llama 3 8B Instruct Abliterated V38-bit8K12 tok/sNot rated
EXAONE 3.5 7.8B Instruct8-bit32K12 tok/sNot rated
Saiga Llama3 8B8-bit8K12 tok/sNot rated
DeepSeek R1 0528 Qwen3 8B8-bit44K12 tok/sNot rated
Gemma 3 12B IT8-bit28K8.0 tok/sText and images
Phi 3.5 Mini Instruct8-bit28K23 tok/sNot rated
Qwen3 4B Thinking 25078-bit72K22 tok/sNot rated

Best fit is the highest quality, up to 8-bit, that leaves the use case’s context. Ranked by our use-case scores, then by Hugging Face downloads. Speeds are rough, plus or minus 25%. The top 60 of 246 are shown.

Image and video models

Image and video models that run on the Mac mini M4 24 GB
ModelBest fitPer outputStrongest at
Stable Diffusion XL 1.016-bit2.7 minImages
Z-Image Turbo8-bit3.9 minImages
FLUX.1 [dev]5-bit8.9 minImages
Qwen-Image 2.14-bit74 minImages
FLUX.3 Klein8-bit5.7 minImages
FLUX.2 [klein] 9B4-bit2.1 minImages
Krea 2 Turbo5-bit8.9 minImages
Stable Diffusion 3.5 Large5-bit39 minImages
MOSS-Image8-bit42 minImages
Wan 2.2 TI2V-5B8-bit6.0 hVideo
MOSS Image Base8-bit42 minImages
HunyuanVideo 1.54-bit31.0 hVideo

Only just too big

Questions

What is the best LLM for the Mac mini M4 24 GB?

Of the open models we track, the strongest that run well on the Mac mini M4 24 GB are Qwen3.8 27B, Muse Glimmer 30B, Gemma 4 26B A4B. 246 text models fit it in all.

Calculated from published specs, not benchmarked. All machines