In this post, we will see what LM Studio is and how we can use it to run local LLMs directly on our machine.
In a previous post, we saw how to run local Large Language Models using Ollama.
Ollama is a very good tool, especially if we want something simple, fast and terminal-oriented.
However, there is another very interesting tool that we can use when we want a more complete local environment, with a nice user interface, model management, configuration options and also a local API server: LM Studio.
But first of all, what is LM Studio?
“LM Studio is a desktop application that allows us to download, configure and run Large Language Models locally on our machine.
We can think of it as a local AI workstation where we can search for models, download them, run them, chat with them and expose them through a local API.
The important point is that everything runs locally.
This means that, when we use a local model with LM Studio, our prompts, our code and our documents are not sent to an external provider like OpenAI, Anthropic or Google.
Obviously, this does not mean that a local model is always better than a cloud model. Cloud models are usually more powerful, more updated and better for complex reasoning tasks.“
LM vs OLLAMA
Before installing LM Studio, let’s make a short comparison with Ollama.
Ollama is probably the easiest way to run local LLMs from the terminal.
For example, with Ollama we can run a model using a command like this:
ollama run llama3
Ollama is perfect when we want a developer-friendly CLI, when we want to automate things, or when we want to run models from scripts or Docker-based environments.
LM Studio, instead, is more focused on giving us a complete desktop experience.
With LM Studio we can search and download models from the UI, configure parameters, manage the context size, load and unload models, use a chat interface and expose a local API server.
Ollama is better if we want a minimal and terminal-first experience.
LM Studio is better if we want a complete UI, more control over model configuration and an easier way to test different models.
Install LM Studio
In my case, I have a Mac with an M4 chip and 24GB of RAM.
This is a good machine for running local LLMs, especially models between 7B and 14B parameters, using quantized versions.
To install LM Studio, we have to go to the official website https://lmstudio.ai/ where we can download the macOS version for Apple Silicon:
From there, we can download the macOS version for Apple Silicon.
After downloading the .dmg file, we install it like a normal macOS application.
Once installed, we can open LM Studio.
LM Studio also includes a CLI called lms.
After opening LM Studio at least once, we can open a terminal and run:
lms --help
If everything is configured correctly, we should see the list of available commands.
For example, we can list the models installed locally with:
lms ls
We can see the models currently loaded in memory with:
lms ps
And we can start the local server with:
lms server start
This is very useful because it allows us to use LM Studio not only from the UI, but also from the terminal and from our applications.
How to download LLMs in LM Studio
After installing LM Studio, we can download models directly from the application.
We open LM Studio, go to the model search section and search for the model we want to try.
In my case, I want to test two models:
Gemma 4 12B QAT
Qwen2.5-Coder 14B
Gemma 4 12B QAT is interesting as a general-purpose model.
QAT means Quantization-Aware Training, so the model is optimized to keep good quality even when running in a quantized version.
Qwen2.5-Coder 14B, instead, is more specific for software development tasks like code generation, code reasoning and code fixing.
Before downloading a model, we have to pay attention to quantization.
Quantization reduces the size of the model and the amount of memory needed to run it locally.
In LM Studio, we can usually find versions like:
Q4
Q5
Q6
Q8
or:
Q4_K_M
Q5_K_M
Q8_0
In general:
Q4 = lighter and faster
Q5 = good balance
Q8 = better quality but heavier
After choosing the model version, we click download, wait until the model is available locally, then load it and start using it from the LM Studio chat interface:
LM Studio is a very useful tool if we want to run local LLMs with a more visual and complete experience compared to a CLI-first tool like Ollama.
With LM Studio, we can search models, download them, load them, test prompts and also expose a local API server that can be used from our applications.
In my opinion, LM Studio is a very good option when we want to experiment with local LLMs in a simple way, without immediately writing code or configuring too many things.
Ollama remains a great choice for terminal-first workflows, but LM Studio is probably better when we want to explore models, compare configurations and quickly test how a model behaves before using it inside a real application.