← Retour au blog
tech 11 May 2026

Running Local Models on a MacBook Pro with 24GB Memory

Discover how to leverage your local machine to run AI models, exploring configurations and models suitable for a MacBook Pro with 24GB memory.

Article inspired by the original source
Running local models on an M4 with 24GB memory ↗ jola.dev

Introduction

In a world where AI is developing at breakneck speed, the idea of running models locally on your own hardware is both fascinating and practical. No need to rely on tech giants for every computational task. Today, we'll explore how you can leverage your MacBook Pro with 24GB of memory to run local AI models.

Why Run Models Locally?

Running AI models locally has several advantages. First, it offers independence from cloud services that can be costly and limiting in terms of privacy. It also allows you to work even without an Internet connection, which is a major asset for some developers.

Hardware and Software Configurations

To begin, you need an adequate software setup. Three main tools stand out: Ollama, llama.cpp, and LM Studio. Each has its quirks, so the choice will depend on your project and personal preferences.

Next, choosing the model is crucial. With 24GB of memory, you can consider models like Qwen 3.5-9B (4b quant) which offers a good balance between performance and memory usage. It can process around 40 tokens per second, which is sufficient for basic development and research tasks.

Configuration Example

Let's take Qwen 3.5-9B as an example. Here are some recommended settings to enable "thinking" mode and perform precise coding tasks:

  • Temperature: 0.6
  • Top_p: 0.95
  • Top_k: 20
  • Min_p: 0.0
  • Presence_penalty: 0.0
  • Repetition_penalty: 1.0

To enable thinking mode, simply add {%- set enable_thinking = true %} to the prompt template in the Inference tab.

Practical Use Case

Consider a web developer using this setup to write code. The model can help generate snippets, suggest optimizations, or even debug existing code. While it's not as performant as a state-of-the-art model, it offers significant flexibility and independence.

Conclusion

Running models locally on a MacBook Pro with 24GB of memory is not only possible but also very practical for certain types of tasks. With the right tools and configurations, you can reduce your dependence on cloud services while maintaining full control over your work environment.

Let's discuss your project in 15 minutes.

local AI models MacBook Pro memory optimization Qwen 3.5-9B LM Studio
Deepthix newsletter · 100% AI · every Monday 8am

An AI agent reads tech for you.

Our AI agent scans ~200 sources per week and ships the best articles to your inbox Monday 8am. Free. One click to unsubscribe.

Visit the newsletter page →

Want to automate your operations?

Let's talk about your project in 15 minutes.

Book a call