Storage CleanerHow much disk space do local LLMs need on a Mac?
Short answer: plan on about 5 GB per 7–8B model, 20 GB per 32B model and 42.5 GB per 70B model at the usual 4-bit quantization. People who try several models typically end up with 30–150 GB: the eight most popular Ollama models together take 114.7 GB.
Size by model class
| Class | Example | Ollama | MLX 4-bit |
|---|---|---|---|
| Small (1–4B) | Llama 3.2 3B | 2.0 GB | 1.8 GB |
| 7–8B | Llama 3.1 8B | 4.9 GB | 4.5 GB |
| 27B | Gemma 3 27B | 17.4 GB | 16.9 GB |
| 32B | Qwen2.5 Coder 32B | 19.9 GB | 18.4 GB |
| 70B | Llama 3.3 70B | 42.5 GB | 39.7 GB |
| 100B+ | gpt-oss 120B / Qwen3 235B | 65.4 GB / 142.2 GB | — |
Ollama: registry manifests. MLX: total file size of the mlx-community repositories on Hugging Face. Collected 2026-09-29. Full list: Ollama model sizes in GB.
Why it adds up faster than you expect
- Every app keeps its own copy. Ollama, LM Studio and Hugging Face / MLX don't share files. Trying Qwen2.5 Coder 32B in Ollama and MLX stores 38.3 GB.
- Several quantizations of one model. Downloading a 4-bit and an 8-bit version roughly triples the space of the 4-bit one alone.
- Models you tried once stay forever. Nothing removes a model you stopped using.
- Hidden locations. Models live in
~/.ollama/models,~/.lmstudio/modelsand~/.cache/huggingface/hub, which Finder doesn't show and macOS counts as System Data.
See what's on your Mac
du -sh ~/.ollama/models ~/.lmstudio/models ~/.cache/huggingface/hub 2>/dev/null
ollama listHow much free space to leave
Keep at least the size of the model you're downloading plus 20 GB free: a download needs its full size, and macOS slows down and can fail updates when the disk is nearly full. On a 256 GB Mac that usually means keeping big models on an external SSD. See is 256 GB enough for programming.
Free space without losing what you use
- Remove models you no longer run: delete Ollama models, LM Studio models, Hugging Face cache.
- Keep one copy of each model, in the app you use most.
- Move large collections to an external drive: OLLAMA_MODELS.
Find every model, in every app. Storage Cleaner lists Ollama, LM Studio and Hugging Face / MLX models together with their real size and when you last used each one, so duplicates and forgotten models are easy to spot and remove, with Undo.
Try Storage Cleaner freeFrequently asked
How much storage for a local LLM on a laptop? One 7–8B model needs about 5 GB. If you plan to try several models, budget 50–100 GB.
Is disk space or memory the limit? Both, differently: disk space limits how many models you can keep; memory limits the largest one you can run. A model generally needs about its file size in free memory to run.
Can I run models from an external SSD? Yes. Loading is a little slower the first time; generation speed is the same once the model is in memory.