Getting Started with LM Studio

What LM Studio Is

LM Studio is a free desktop application for discovering, downloading, and running open-source AI models on your own machine. Where Ollama is fundamentally a CLI tool, LM Studio gives you a graphical interface for everything — browsing a model catalogue, downloading models with a progress bar, loading them with a click, chatting in a clean UI, and managing your local model library without ever opening a terminal.

LM Studio runs on macOS 14+ (Apple Silicon), Windows, and Linux. Current version as of mid-2026 is 0.4.23. Under the hood it uses llama.cpp for CPU/NVIDIA/AMD inference and Apple’s MLX framework for Apple Silicon — the same engines Ollama uses, but wrapped in a full desktop experience.

Installation

Download LM Studio from lmstudio.ai. On macOS, open the .dmg and drag LM Studio to your Applications folder. On Windows, run the installer. On Linux, use the provided AppImage.

Launch LM Studio. No account required. No API key needed. The application opens directly to the main interface.

The Interface: Three Main Tabs

LM Studio is organised around three areas you will use constantly:

Discover (the search icon) — Browse and download models. This connects to Hugging Face and shows models filtered for LM Studio compatibility. You can search by name, filter by parameter count, and see file sizes for each quantization level before downloading. On Apple Silicon, models appear with a “MLX” badge when an MLX-optimised version is available — always choose that over GGUF for the best performance on M-series Macs.

Chat (the speech bubble icon) — Your main workspace. Load a model, type messages, and get responses. The right panel shows generation settings you can adjust live: temperature, context length, top-p, and others.

My Models (the folder icon) — All models you have downloaded. Shows file size, quantization level, and format. Right-click any model to delete it and free disk space.

Downloading Your First Model

Go to the Discover tab. Search for llama-3.3-8b-instruct. You will see results from several publishers. Choose the one from lmstudio-community — these are tested, recommended builds curated for LM Studio users.

Click the model name. A panel opens showing available quantizations and their file sizes. For a first download on a machine with 8 GB VRAM, select Q4_K_M (around 4.7 GB). Click the download button next to it. A progress bar appears — the download runs in the background and you can navigate to other tabs.

On Apple Silicon, search for llama-3.3-8b-instruct-mlx instead and download the MLX version for significantly better performance.

Your First Chat

Once the download is complete, go to the Chat tab. In the top bar, click the model selector and choose the model you just downloaded. LM Studio loads it into memory — you will see a VRAM usage bar fill as the model loads. When it turns green, the model is ready.

Type a message in the input box at the bottom and press Enter. The model generates a response in the chat window. Token generation speed appears in small text below the response — on a 8 GB VRAM card you should see 20–40 tokens/sec; on Apple Silicon M3+ with MLX you should see 60–100+ tokens/sec for an 8B model.

The right settings panel lets you adjust temperature, context length, and other parameters live. Changes take effect on the next message — you do not need to reload the model.