How to Use Ollama and Qwen on a Mac: A Complete Beginner’s Guide

On a supported Mac, install Ollama and run ollama run qwen3.5:4b in Terminal to download Qwen 3.5 4B and start a chat. Check macOS compatibility and free storage first.

Not every Mac will produce responses at the same speed, however. A model's download size is also not the same as the memory it consumes while running. This guide takes you from the compatibility check to installation, your first prompt, switching to a lighter model, managing downloads, and fixing common setup problems without assuming previous command-line experience.

What Ollama and Qwen do

Ollama is an application and command-line interface for downloading and running supported AI models on a computer. Qwen is the model family that generates the responses. A useful mental model is that Ollama supplies the local runtime, while Qwen supplies the language model inside it.

Qwen is developed by the Qwen Team at Alibaba Group. The official repository records the release of Qwen 3.5 0.8B, 2B, 4B, and 9B models in March 2026. Qwen's official repository

You need an internet connection to download Ollama and the model files. After a local model has been downloaded, the model can process a chat on the Mac itself. This distinction matters because Ollama also offers cloud-hosted models, which are not the same as a downloaded local model.

Check your Mac and free storage first

Ollama's current Mac requirements specify macOS Sonoma 14 or newer. Apple M-series Macs have CPU and GPU support, while x86-based Intel Macs use the CPU. An Intel Mac can therefore be supported, but you should expect its experience to differ from an Apple Silicon Mac. Ollama's official macOS documentation

Open System Settings, choose General, and check Storage before downloading a model. The recommended starting point in this guide, qwen3.5:4b, is listed as a 3.4GB model file in the Ollama library. Leave additional room for the Ollama application, temporary data, and any models you may try later. Official Qwen 3.5 4B library page

The listed 3.4GB is the model file size, not a promise about runtime RAM or unified-memory usage. Actual performance depends on the Mac model, installed memory, other active applications, and the length of your conversation.

Install Ollama on macOS

Download the Mac DMG from Ollama's official site. Open it, then move the Ollama application into the system Applications folder. This is the installation method recommended in the official macOS documentation. When Ollama starts, it checks whether the ollama command is available in your PATH. If it is missing, the app may ask permission to create a link in /usr/local/bin. Read the prompt and allow it so Terminal can find the command. Ollama's macOS installation instructions

Open Terminal after installation and run:

ollama

If the interactive menu opens, the command-line setup is ready. Ollama's current Quickstart uses the same ollama command to open the menu, where you can run a model or launch a supported integration. Ollama Quickstart

Download Qwen 3.5 4B and start chatting

For a first attempt, use the 4B variant as a clear middle starting point. Enter this command in Terminal:

ollama run qwen3.5:4b

The first run downloads the model, so the wait depends on your connection. The official Qwen 3.5 4B library page provides this exact command for the model. Qwen 3.5 4B run example

When a prompt marker such as >>> appears, try a task that does not require live web information, such as: "Make three bullets: read at 9 a.m., lunch at noon, shopping at 3 p.m." Starting with a short request makes it easier to judge response speed. To leave the interactive chat, enter:

/bye

The Ollama Quickstart also documents /bye as the command for leaving a chat. Ollama Quickstart

Switch to a smaller model if the Mac feels slow

If 4B responds too slowly or makes other applications feel unresponsive, try a smaller model before changing advanced settings. Ollama lists qwen3.5:0.8b as a 1.0GB model file and qwen3.5:2b as a 2.7GB model file. Qwen 3.5 0.8B Qwen 3.5 2B

Run either model with the same pattern:

ollama run qwen3.5:2b

If your Mac has more headroom and you want to try a larger option, qwen3.5:9b is listed as a 6.6GB model file. Qwen 3.5 9B

These download sizes should not be treated as exact memory requirements. A sensible sequence is to start with 4B, move down to 2B or 0.8B if it is uncomfortable, and consider 9B only when your Mac has enough storage and memory headroom. Smaller models reduce local resource demand, but their answers may also be less capable on difficult tasks.

List, inspect, stop, and remove models

Four commands cover the model-management tasks most beginners need:

ollama ls
ollama ps
ollama stop qwen3.5:4b
ollama rm qwen3.5:4b

ollama ls lists downloaded models. ollama ps lists models currently loaded into memory. ollama stop stops a loaded model, while ollama rm removes the downloaded model files. Removing a model means you will need to download it again if you want to use it later, so check the model name before running rm. All four operations are documented in the official CLI reference. Ollama CLI Reference

The PROCESSOR column shown by ollama ps can also indicate whether a model is loaded on the CPU, GPU, or split between both. This is useful when you are checking why performance differs between machines. Ollama FAQ

Troubleshoot the most common setup problems

If Terminal reports command not found: ollama, start the Ollama application once and look for the prompt that creates the CLI link. When Ollama is installed outside Applications, its CLI binary or a symbolic link to that binary must be on your PATH. The official macOS page identifies the CLI inside Ollama.app/Contents/Resources/ollama. Ollama macOS troubleshooting information

If a download does not progress, check the internet connection and available storage. If generation is uncomfortably slow, close other demanding applications and try qwen3.5:2b or qwen3.5:0.8b. Use ollama ps to inspect how the model is loaded rather than assuming that the GPU is active.

Ollama stores its Mac configuration under ~/.ollama. The documented log directory is ~/.ollama/logs, including app.log and server.log. Logs can help identify a recurring error, but remove passwords, API keys, prompts, and other sensitive content before sharing log excerpts. Ollama's macOS documentation

Understand local privacy and the limits of the model

Ollama states that it does not see prompts or data when you run a model locally. Cloud-hosted models are different: prompts and responses are processed to provide the cloud service. Do not assume that every model reached through Ollama has the same data path; check whether the model is local or has a :cloud tag. Ollama's data-handling FAQ

Local execution can keep a downloaded model's chat processing on your Mac, but it also consumes your storage and memory and depends on your hardware. It does not make the answer automatically correct. A local Qwen model may produce outdated or inaccurate statements, and a basic local chat does not automatically know current web information. Verify important facts with primary sources, and avoid entering passwords, API keys, or unnecessary personal information even in a local workflow.

The benefit is control and a low-friction way to experiment. The tradeoff is that you are responsible for choosing a model your Mac can handle, maintaining its files, and checking the output before relying on it.

Record the runtime before changing models

First record ollama --version and the downloaded tags in ollama ls. During generation, inspect ollama ps in another terminal. Compare 4B and 2B using the same short input in fresh conversations. Separate first-download time from response generation.

An illustrative task is “Make a two-column table of item and quantity: three apples and two oranges.” Check preserved names and quantities and the absence of invented details. Record answer correctness separately from Mac responsiveness. If performance is uncomfortable, stop the model and try a smaller tag. This article does not claim measured physical-device speed or memory usage.

Summary

The shortest path to using Ollama and Qwen on a Mac is to confirm macOS and free storage, install the Ollama application, and run ollama run qwen3.5:4b. If 4B is too heavy, step down to 2B or 0.8B. Use ollama ls, ps, stop, and rm to keep models under control, and remember that local and cloud-hosted models have different data paths.

One short prompt is enough to begin: finding the model that feels comfortable on your own Mac is what turns local AI from an intimidating setup into a useful everyday tool.