Gemma 4 is Google's free open model, and it runs on any Apple Silicon Mac with nothing sent to Google. On a 16 GB Mac, install the 12B size (a 7.7 GB download on a Mac), or the faster E4B, which is a tighter fit because Ollama gives Macs a 9.5 GB build of it; on a 32 GB Mac, the 26B mixture-of-experts or the 31B model (16 to 20 GB). One command in Ollama gets you started: ollama run gemma4. It reads images as well as text, can call tools for agents and has a thinking mode for harder questions.
Key facts
- Sizes: E2B, E4B, 12B, 26B (mixture of experts, about 4B active) and 31B (dense).
- Downloads in Ollama: E2B 4.6 to 7.5 GB, E4B 6.6 to 9.5 GB, 12B 7.7 to 8.0 GB, 26B 16 to 19 GB, 31B 19 to 20 GB.
- Context: 128K tokens on E2B and E4B, 256K on 12B, 26B and 31B.
- Can do: text and images on every size; audio on E2B and E4B; tool calling and a configurable thinking mode.
- Cost: free to download and run. Ollama lists it under the Apache 2.0 licence, which allows commercial use.
On this page
- Which Gemma 4 size fits your Mac?
- How to run Gemma 4 on a Mac
- What Gemma 4 is good at
- Gemma 4 vs Qwen3.8 and gpt-oss
- Is it private?
- Frequently asked questions
Which Gemma 4 size fits your Mac?
Memory decides it. The model has to fit in your Mac's memory with room left for macOS, your apps and the conversation, so pick a download well under your total.
| Your Mac's memory | Gemma 4 size | Ollama command | Download |
|---|---|---|---|
| 8 GB | E2B | ollama run gemma4:e2b | 7.5 GB on a Mac, too big for 8 GB: see the 8 GB picks |
| 16 GB | E4B (the default) | ollama run gemma4 | 9.5 GB on a Mac, 6.6 GB elsewhere |
| 24 GB | 12B | ollama run gemma4:12b | 7.7 GB |
| 32 GB or more | 26B (fast) or 31B (best quality) | ollama run gemma4:26b | 16 to 20 GB |
The 26B is a mixture-of-experts model: only about 4B of its parameters work on each word, so it answers faster than the 31B for a similar download. The 31B is dense, slower and the strongest of the family. Not sure what fits? Try the local LLM memory calculator.
How to run Gemma 4 on a Mac
- Install Ollama from ollama.com, or LM Studio if you prefer a chat window. See Ollama vs LM Studio.
- Open Terminal and type
ollama run gemma4(or the size from the table). The first run downloads the model. - Chat in the Terminal, or connect a front end such as Open WebUI or Msty.
- On a Mac, try the MLX versions (for example
gemma4:26b-mlx), which Ollama lists for every size and which use Apple's own machine-learning framework.
In LM Studio, search for Gemma 4 in the Discover tab and download the size that matches your memory. For more commands, see the Ollama commands cheat sheet.
What Gemma 4 is good at
- Everyday drafting, summaries and questions, which is why my Ollama models list makes it the default for most Macs.
- Reading images and screenshots, on every size.
- Agents: its tool calling works with local agent apps that expect a model to call functions. See agent skills on local models.
- Long documents on the 12B, 26B and 31B, with a 256K-token context window, if you have the memory for a long conversation.
Like every local model, it trails the biggest cloud models on hard reasoning and very long, multi-step coding jobs.
Gemma 4 vs Qwen3.8 and gpt-oss
- For coding and agents on 32 GB: Qwen3.8 27B is my first pick.
- For step-by-step reasoning on 24 GB: gpt-oss 20B.
- For everyday work and images on 16 GB: Gemma 4 E4B.
Many people keep two models installed and switch by job. See the best local LLMs for a Mac for picks by memory and task.
Is it private?
Yes, when it runs locally: the model and your conversations stay on your Mac, and it works with the Wi-Fi off. Avoid the tags marked cloud, such as gemma4:cloud, which run on Ollama's servers, so your prompts leave the machine. See private AI.
Frequently asked questions
Can Gemma 4 run on a Mac with 16 GB?
Yes. On a Mac, Ollama downloads a 9.5 GB build of the default E4B size (ollama run gemma4), which fits a 16 GB Mac if you close other apps; it answered at 87 tokens a second in my tests. The 12B is a 7.7 GB download on a Mac and leaves more room.
Which Gemma 4 model is best for a 32 GB Mac?
The 26B mixture-of-experts model for speed, or the 31B dense model for the best quality. Both are 16 to 20 GB downloads with a 256K-token context window.
Is Gemma 4 free?
Yes. It costs nothing to download or run on your own Mac, and the Ollama library lists it under the Apache 2.0 licence, which allows commercial use.
Does Gemma 4 work offline?
Yes. Once downloaded, the local versions run with no internet connection. Only the tags marked cloud need Ollama's servers.
Sources
- Ollama, gemma4 in the Ollama library (read 21 November 2026)
- Ollama models list, pull counts and licences read 25 October 2026
Sizes change when new versions are published. Check the Ollama library before you download. Spotted something out of date? Tell me.



