June 11, 2026·5 min read

Why Is Ollama Taking So Much Disk Space?

Downloading a few local models can fill your SSD faster than expected. Here is what is actually happening — and why it is harder to clean up than it looks.

Ollamadisk spacelocal AIcleanup

You installed Ollama. You pulled a model or two. Maybe you ran a few comparisons. Now your Mac storage indicator is alarming and you are wondering what happened.

Here is what happened: local AI models are enormous, they accumulate fast, and Ollama makes it very easy to download more without making it easy to see how much space you have already committed.

How fast it adds up

A single 8B parameter model in common quantization is roughly 5 GB. A 70B model is 40+ GB. If you have tested three or four models — which takes maybe an afternoon — you can be looking at 60 to 80 GB of model storage before you have noticed anything wrong.

The problem gets worse because:

  • The same model may exist in multiple quantizations. Q4, Q8, and fp16 versions of the same model are different files. Pull all three to compare quality and you just tripled the footprint.
  • Multiple tools share the same models without knowing about each other. If you use both Ollama and LM Studio, you may have downloaded the same underlying model twice — in different formats, in different hidden directories.
  • Old experiments do not clean themselves up. A model you pulled in January to test a feature is still sitting on your drive in July.

Want to skip the hidden-folder hunt?
LLM Cleaner scans your Mac for local AI models, caches, indexes, and project memory — then shows what you can review, reveal, export, or safely move to Trash.

Scan with LLM Cleaner

Why you cannot just search for large files

If you use Finder or a standard disk analyzer to find large files, you will probably find the model blobs. But you will not know which model they belong to, whether they are still referenced, or whether deleting them will silently break something.

Ollama model files are stored with cryptographic hash names in a hidden directory. A standard file browser shows you raw file sizes but gives you no context about what each file is for.

The cleanup approach that actually works

Effective Ollama cleanup requires three things that a normal disk cleaner cannot provide:

  1. Cross-referencing blobs with manifests so you know which files belong to which model
  2. Identifying orphaned blobs that are no longer referenced by any model
  3. Seeing your total Ollama footprint alongside other AI tools so you can prioritise what to remove

Without those three things, you are deleting blind.

Local AI storage is messy. Cleanup should not be.
Find Ollama, LM Studio, Hugging Face, Cursor, Windsurf, Claude Code, Whisper, and loose model files in one place.

Get Founder License — $19
More articles
June 10, 2026·5 min read

Where Does Ollama Store Models on Mac?

Ollama models are powerful, but they quietly take up gigabytes inside hidden folders most users never open. Here is what to know before you start deleting.

June 12, 2026·5 min read

Where Are Hugging Face Models Stored on Mac?

You may have downloaded models through Python, Transformers, Whisper, or Sentence Transformers — and never realised Hugging Face kept them locally. Here is what to know before cleaning up.