Storage CleanerResearch · updated September 26, 2026
Mac developer storage statistics (2026)
We measured how much disk space the most common developer and local-AI tools take on a Mac in 2026, and collected the key figures around them from primary sources: Ollama, Hugging Face, npm, PyPI, Docker Hub, Apple and the Stack Overflow Developer Survey. Every number below links to where it comes from.
Key takeaways
- Llama 3.3 70B takes 42.5 GB of disk in Ollama. (Ollama)
- Eight popular Ollama models together take 114.7 GB. (Ollama)
- Llama 3.1 has been downloaded 119.9M times from Ollama; DeepSeek-R1 93.2M. (Ollama)
- Keeping one 32B coding model in both Ollama and MLX costs 38.3 GB. (Ollama, Hugging Face)
- The mlx-community organisation on Hugging Face publishes 5,605 models for Apple silicon. (Hugging Face)
- Xcode 26.5 plus one iOS simulator runtime take 13.6 GB before you build anything. (measured)
- Each iOS simulator runtime is about 8.5 GB. (measured)
- A new Next.js project installs 369 packages, at least 429 MB. (npm)
- A new Vite + React project is 70 MB, about six times smaller. (npm)
- A 256 GB MacBook gives macOS 245.1 GB of usable capacity. (measured)
- On MacBook Neo, 512 GB instead of 256 GB costs $100 ($699 vs $799); MacBook Air now starts at 512 GB. (Apple)
- 32.9% of professional developers use macOS as their primary work OS. (Stack Overflow 2025)
- 84% of developers use or plan to use AI tools, up from 76%. (Stack Overflow 2025)
How much disk space do local AI models use on a Mac?
Popular local AI models take 2–43 GB each in Ollama's default quantisation, and eight of the most-used ones total 114.7 GB. (Ollama registry)
How popular are these models?
| Model | Ollama downloads |
|---|---|
llama3.1 | 119.9M |
deepseek-r1 | 93.2M |
llama3.2 | 84.4M |
gemma3 | 40.7M |
qwen3 | 38M |
qwen2.5-coder | 21.8M |
gpt-oss | 13.2M |
llama3.3 | 4.2M |
mistral-small | 3.1M |
Download counts as shown on each model's page at ollama.com on September 26, 2026.
Does the same model take space twice?
Yes. Ollama, Hugging Face/MLX and LM Studio each keep their own copy, so Qwen2.5 Coder 32B takes 19.9 GB in Ollama and 18.4 GB as an MLX download: 38.3 GB for one model. (Ollama, Hugging Face) The same families as 4-bit MLX models (total repository size):
Hugging Face's mlx-community organisation alone publishes 5,605 models converted for Apple silicon. (Hugging Face) See how to delete Ollama models and how to clear the Hugging Face cache.
How much space does Xcode use?
Xcode 26.5 takes 5.1 GB, and one iOS simulator runtime adds 8.5 GB: 13.6 GB before you build anything. (measured)
| Item | Size | How measured |
|---|---|---|
| Xcode 26.5 app | 5.1 GB | du on /Applications/Xcode.app |
| iOS 26.5 simulator runtime | 8.5 GB | xcrun simctl runtime list |
| Total before building anything | 13.6 GB |
DerivedData, iOS DeviceSupport (3–6 GB per iOS version of each connected device) and old runtimes come on top. See how to clear Xcode's leftovers.
How big is node_modules for a new JavaScript project?
A new Next.js project installs 369 packages taking at least 429 MB, while a Vite + React project installs 25 packages and 70 MB. (npm registry)
| Starter stack | Packages | node_modules size |
|---|---|---|
| Next.js 16.3.6 (TypeScript, Tailwind, ESLint) | 369 | at least 429 MB |
| Expo SDK 57 (React Native 0.87.1) | 496 | at least 283 MB |
| Vite 8.3.1 + React (TypeScript) | 25 | 70 MB |
"At least" because a few packages don't publish their size (12 for Next.js, 43 for Expo). These packages are downloaded constantly: in the month to 2026-09-24, npm served next 197 million times, react 627 million, vite 643 million and typescript 1.0 billion times. (npm downloads API) See how to delete every node_modules folder.
How big are Python machine-learning packages on macOS?
The PyTorch wheel for Apple silicon is 127 MB to download and TensorFlow's is 223 MB; installed, they take more. (PyPI)
| Package | Version | macOS arm64 download |
|---|---|---|
tensorflow | 2.21.0 | 223 MB |
torch | 2.14.0 | 127 MB |
opencv-python | 5.0.0.93 | 48 MB |
jupyterlab | 4.6.4 | 17 MB |
transformers | 5.17.0 | 12 MB |
numpy | 2.5.3 | 12 MB |
mlx | 0.32.2 | 1 MB |
mlx-lm | 0.31.3 | 0 MB |
Each project's virtual environment gets its own copy, and pip and uv cache every wheel they download. See how to clear package caches.
How big are common Docker images?
Official images for Node.js and Python are about 400 MB each to download for Apple silicon, compressed. (Docker Hub)
| Image (linux/arm64) | Compressed download | All-time pulls |
|---|---|---|
node:22 | 400 MB | 7.0 billion |
python:3.12 | 400 MB | 9.3 billion |
postgres:17 | 160 MB | 11.6 billion |
mysql:8.4 | 234 MB | 5.2 billion |
redis:7 | 43 MB | 11.3 billion |
ubuntu:24.04 | 29 MB | 10.1 billion |
Images take more space once unpacked, and every version you pull stays until you prune it. See how to shrink Docker's disk.
How much usable space does a 256 GB Mac have?
The 256 GB MacBook we measured has an SSD of 251.0 GB, of which macOS gets a 245.1 GB container, the capacity Finder shows. (measured) Apple counts 1 GB as 1 billion bytes and notes that "actual formatted capacity" is less. (Apple) macOS, its apps and System Data then take their share before any of the tools above.
Apple's current 256 GB Mac is MacBook Neo at $699; the 512 GB model is $799. (Apple) MacBook Air starts at $1,299 with 512 GB (options up to 4 TB). (Apple) Xcode with one runtime (13.6 GB) and three mid-size models (about 50 GB) fill a large part of a 256 GB Mac before any projects.
How many developers use a Mac?
32.9% of professional developers use macOS as the primary operating system they work in, and 32.7% for personal use. (Stack Overflow Developer Survey 2025)
How many developers use AI tools?
84% of developers are using or planning to use AI tools in their development process, up from 76% the year before. (Stack Overflow Developer Survey 2025) Running models locally is where the disk space goes.
See your own numbers. Storage Cleaner scans your Mac for every item in this report, shows when you last used each one, and cleans to the Trash with Undo. Scanning is free.
Try Storage Cleaner freeMethod and sources
- Ollama model sizes: sum of layer sizes in each model's manifest in Ollama's public registry (default tag). Download counts: each model's page on ollama.com.
- Hugging Face: total size of all files in each mlx-community repository and the organisation's model count, from the Hugging Face API.
- npm: npm resolved a lockfile for the latest versions of each stack (nothing installed); we summed each package's published unpacked size, counting only packages that install on macOS arm64. Downloads: the npm downloads API, last 30 days.
- PyPI: size of the newest macOS arm64 wheel (or the universal wheel) from the PyPI JSON API.
- Docker Hub: compressed linux/arm64 size for each tag and all-time pull count, from the Docker Hub API.
- Measured: Xcode 26.5 and disk capacity on a 256 GB MacBook Air (
du,xcrun simctl,diskutil). - Apple and Stack Overflow: read from the linked pages on September 26, 2026. Prices are US prices.
Registry data collected September 26, 2026 (2026-09-26T12:25+00:00). Sizes in decimal units (1 GB = 109 bytes), as macOS reports them. We update this page every quarter.
Using this data
You're welcome to use these figures, with a link to this page (CC BY 4.0). Suggested citation:
Storage Cleaner, "Mac Developer Storage Statistics 2026", storagecleaner.dev/research/developer-disk-space-2026, September 26, 2026.
Download the data: developer-disk-space-2026.json. Spotted an error? Email [email protected].