Skip to main content
PromptQuorumBuilt for humans. Structured for AI.
Home/Local LLMs
Getting Started

Install LM Studio: GUI Setup for macOS, Windows & Linux

·7 min readBy · Founder of PromptQuorum · Discovery engine for open-weight & open-source AI

LM Studio is a desktop application that lets you browse, download, and run local LLMs through a graphical interface -- no terminal commands required. It runs on macOS, Windows, and Linux, and includes a built-in chat UI and an OpenAI-compatible local server.

Key Takeaways

  • Download LM Studio from lmstudio.ai -- available for macOS (Apple Silicon), Windows, and Linux (AppImage).
  • Requirements: macOS 14.0+ on Apple Silicon, Windows x64 (AVX2) or ARM64, or Ubuntu 20.04+; 16 GB RAM recommended. Apple Silicon Macs use GPU acceleration by default.
  • The built-in model browser searches Hugging Face directly -- download GGUF models without leaving the app.
  • LM Studio includes a built-in chat UI and a local OpenAI-compatible server on port 1234.
  • Best for: beginners who prefer a GUI, users who want to compare multiple models side-by-side, and anyone who wants a complete package without terminal commands.

📍 In One Sentence

LM Studio is a desktop application for macOS, Windows, and Linux that lets you browse, download, and chat with many GGUF-quantized models from Hugging Face (models with an unsupported architecture may not load) through a graphical interface, with no terminal commands required and a built-in OpenAI-compatible local server. According to the vendor, use is free, but the terms can change.

💬 In Plain Terms

If you don't want to use a command line, LM Studio is the point-and-click way to run AI models on your own computer. Download the app, search for a model by name inside it, click download, and start chatting -- the whole thing takes under 5 minutes and includes a built-in server so other apps can talk to it too.

LM Studio is a desktop application that lets you browse, download, and run local LLMs through a graphical interface -- no terminal commands required. It runs on macOS, Windows, and Linux, and includes a built-in chat UI and an OpenAI-compatible local server. LM Studio supports many GGUF-quantized models from Hugging Face; models with an unsupported architecture may not load.

Install LM Studio: GUI Setup for macOS, Windows & Linux

Download LM Studio and Install It: One Step per OS

Download LM Studio from the official LM Studio download page, then follow the one install step for your system below. It needs Apple Silicon with macOS 14.0 or newer, a Windows x64 (AVX2) or ARM64 PC, or Ubuntu 20.04 or newer, and 16 GB of RAM is recommended (see the requirements table). According to the vendor (lmstudio.ai/blog/free-for-work), use is free, but the terms can change.

  • macOS: open the downloaded .dmg file and drag LM Studio into the Applications folder.
  • Windows: run the downloaded installer and follow the prompts.
  • Linux: make the downloaded AppImage executable with `chmod +x LM-Studio-*.AppImage`, then run it.

Which Model Fits Your GPU

Match the model to your GPU memory: 8 GB VRAM runs 7–8B models, 12 GB runs 14B, 16 GB runs 24B, 24 GB runs 27–32B, and a 70B model needs about 40 GB. Sizes are approximate estimates for Q4_K_M quantization from the PromptQuorum local LLM hardware guide, which also covers GPUs, prices and speeds. On Apple Silicon, unified memory counts as VRAM.

Your VRAM
Models that fit
Examples
Size at Q4_K_M
8 GB7–8BLlama 3.1 8B, Qwen3 8B~5 GB
12 GB14BQwen3 14B~9 GB
16 GB24BMistral Small 3.1 24B~14 GB
24 GB27–32BQwen3.6 27B, DeepSeek-R1 32B~16–19 GB
40 GB+70BLlama 3.3 70B~40 GB

Estimates, not measurements by PromptQuorum. See the VRAM guide for the formula behind the sizes.

What Is LM Studio?

LM Studio is a desktop application for running local LLMs. It provides a graphical model browser, a built-in chat interface, and a local API server -- all in one app. Under the hood, it runs GGUF models with llama.cpp (and with MLX on Apple Silicon), an open-source engine that Ollama also supports as a backend.

The key difference from Ollama is that LM Studio is entirely GUI-driven. You browse and download models through the app interface, start chats with a click, and manage model settings with sliders rather than configuration files.

According to the vendor (lmstudio.ai/blog/free-for-work), LM Studio has been free for personal and work use since July 2025. It is developed by Element Labs, Inc. As of 2026, it supports NVIDIA CUDA, AMD ROCm, and Apple Metal acceleration.

What Are the System Requirements for LM Studio?

Spec
Minimum
Recommended
Operating SystemmacOS 14.0+, Windows x64 (AVX2) or ARM64, Ubuntu 20.04+Ubuntu 22.04 (newer: less tested)
RAMNo fixed minimum (8 GB Macs: small models)16 GB or more
StorageApp plus space for each model50 GB+ free for several models
GPU (optional)Optional (Windows: 4 GB+ dedicated VRAM)Apple Silicon, or NVIDIA/AMD with enough VRAM

How Do You Download and Install LM Studio?

  1. 1
    Go to lmstudio.ai and click the download button for your OS.
  2. 2
    macOS: Open the .dmg file and drag LM Studio to Applications. On first launch, approve the security prompt in System Preferences → Privacy & Security.
  3. 3
    Windows: Run the LM-Studio-Setup.exe installer. LM Studio installs to %LOCALAPPDATA%\LM-Studio.
  4. 4
    Linux: Download the .AppImage file. Make it executable with `chmod +x LM-Studio-*.AppImage` and run it. No system installation required.
  5. 5
    On first launch, LM Studio shows a welcome screen and prompts you to download a model.
LM Studio installation flow: download from lmstudio.ai, install on macOS, Windows, or Linux, search Hugging Face for a GGUF model, pick Q4_K_M quantization (~4.5 GB), and start chatting in 5-30 seconds.
LM Studio installation flow: download from lmstudio.ai, install on macOS, Windows, or Linux, search Hugging Face for a GGUF model, pick Q4_K_M quantization (~4.5 GB), and start chatting in 5-30 seconds.

How Do You Find and Download a Model in LM Studio?

Use the Search tab (magnifying glass icon in the left sidebar) to find models:

  1. 1
    Click the Search tab in the left sidebar.
  2. 2
    Type a model name -- for example "llama 3.1" or "phi-3 mini".
  3. 3
    LM Studio shows matching GGUF models from Hugging Face with file sizes and quantization options.
  4. 4
    Select a quantization level. For 8 GB RAM: choose Q4_K_M (~4.5 GB for a 7B model). For 16 GB RAM: Q5_K_M or Q6_K offer better quality.
  5. 5
    Click the download arrow. Progress shows in the Downloads tab.
RAM-based quantization guide for LM Studio: 8 GB RAM fits Q4_K_M (~4.5 GB, small quality loss that varies by model); 16 GB RAM fits Q5_K_M or Q6_K (~5.7-6.5 GB, near-lossless) with headroom for 8K+ token context.
RAM-based quantization guide for LM Studio: 8 GB RAM fits Q4_K_M (~4.5 GB, small quality loss that varies by model); 16 GB RAM fits Q5_K_M or Q6_K (~5.7-6.5 GB, near-lossless) with headroom for 8K+ token context.

How Do You Start Chatting with a Model in LM Studio?

  1. 1
    Click the Chat tab (speech bubble icon) in the left sidebar.
  2. 2
    At the top of the chat window, click the model selector dropdown and choose your downloaded model.
  3. 3
    LM Studio loads the model into memory -- this takes 5-30 seconds depending on model size and hardware.
  4. 4
    Type your message in the input field at the bottom and press Enter or click Send.
  5. 5
    The model's response streams token by token. Generation speed appears in the status bar at the bottom of the window.

How Do You Adjust Model Settings in LM Studio?

The right panel in the Chat tab exposes key inference parameters:

  • Temperature (default 0.8): controls response randomness. Lower values (0.1-0.4) produce more focused, predictable output. Higher values (0.8-1.2) produce more varied, creative output.
  • Context Length (default 4096 tokens): the maximum conversation history the model can process. Longer context uses more RAM. Most 7B models support 4096-8192 tokens.
  • GPU Layers (macOS/Linux/Windows with GPU): how many model layers to offload to the GPU. Set to max for fastest inference if your GPU has enough VRAM.
  • System Prompt: a persistent instruction prepended to every conversation. Use this to set the model's role or behavior.

How Do You Enable the LM Studio Local Server?

LM Studio includes a local server that mimics the OpenAI API. Many applications built for the OpenAI API can use your local model through this server, though compatibility is not guaranteed for every feature:

  1. 1
    Click the Local Server tab (the "<->" icon) in the left sidebar.
  2. 2
    Select a model in the model dropdown at the top.
  3. 3
    Click "Start Server". The server starts on http://localhost:1234.
  4. 4
    Your application should set `base_url = "http://localhost:1234/v1"` and any string as the API key (the server accepts any value).

Connect to LM Studio via Python

python
from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:1234/v1",
    api_key="not-needed"
)

response = client.chat.completions.create(
    model="local-model",
    messages=[{"role": "user", "content": "What is a local LLM?"}]
)
print(response.choices[0].message.content)

LM Studio vs Ollama: Which Should You Use?

Factor
LM Studio
Ollama
InterfaceGraphical desktop appTerminal + API
Model sourceHugging Face (many GGUF models)Ollama library (curated)
API portlocalhost:1234localhost:11434
Model managementGUI browser with file size infoCLI commands (ollama pull, list, rm)
AutomationGUI-first; local API and lms CLI availableCLI-first (scripting, Docker, CI)
Best forBeginners, GUI users, model explorationDevelopers, automation, server deployments

Troubleshooting Common LM Studio Issues

LM Studio says "Not enough memory to load model"

The model requires more RAM than is available. Close other applications to free memory, or select a smaller quantization (Q3_K_S instead of Q4_K_M). As a rule: multiply the model file size by 1.2 to estimate the RAM needed. A 4.5 GB file needs ~5.4 GB free RAM.

The model is generating very slowly (under 5 tokens/sec)

The model is running entirely on CPU. Check GPU Layers in the right panel -- if it shows 0, your GPU is not being used. On macOS, LM Studio enables Metal (GPU) automatically for Apple Silicon. On Windows/Linux with NVIDIA, ensure your driver is up to date and increase GPU Layers to the maximum value shown.

I cannot find a specific model in the LM Studio search

LM Studio searches Hugging Face for GGUF files. If a model is not appearing, try searching by the Hugging Face repository name directly (e.g., "bartowski/Llama-3.1-8B-Instruct-GGUF"). Some newer models may not be indexed yet.

The local server returns "model not found" errors

A model must be loaded in the Local Server tab before the server can respond. Open the Local Server tab, select a model from the dropdown, and click Start Server. The model name in API requests can be any string -- LM Studio uses whichever model is currently loaded.

How do I download LM Studio?

Go to lmstudio.ai and click the download button for your OS. Available for macOS (Apple Silicon), Windows, and Linux (AppImage).

What are the minimum requirements for LM Studio?

Apple Silicon with macOS 14.0 or newer, Windows x64 (AVX2) or ARM64, or Ubuntu 20.04 or newer; 16 GB RAM is recommended (8 GB Macs only for smaller models). No GPU required -- Apple Silicon Macs and NVIDIA/AMD GPUs are supported for acceleration.

How do I find and download models in LM Studio?

Click the Search tab (magnifying glass) in the sidebar, search for a model name (e.g., "llama 3.1"), select a quantization level (Q4_K_M for 8GB RAM), and click the download arrow.

What quantization should I use in LM Studio with 8GB RAM?

Q4_K_M is the recommended quantization for 8GB RAM systems. It gives the best balance of model quality and memory usage for 7B models (~4.5GB file size).

Does LM Studio include an OpenAI-compatible API?

Yes. Enable the Local Server tab in LM Studio to start an OpenAI-compatible API at http://localhost:1234. Many OpenAI SDK apps can connect using this URL as the base_url, but compatibility is partial, so test the features you rely on.

How is LM Studio different from Ollama?

LM Studio is GUI-first: browse models, manage settings, and chat through a visual interface. Ollama is CLI-first: faster to set up for developers, but requires terminal commands. LM Studio runs GGUF models with llama.cpp (and MLX on Apple Silicon); Ollama also supports llama.cpp as a backend.

Can I use LM Studio on Linux?

Yes. Download the .AppImage file from lmstudio.ai. Make it executable: chmod +x LM-Studio-*.AppImage and run it. No system installation is needed -- it runs as a portable app.

Is LM Studio free?

According to the vendor (lmstudio.ai/blog/free-for-work), LM Studio has been free for personal and work use since July 2025. It is developed by Element Labs, Inc.; only the separate Enterprise/Teams tier is paid. Terms can change, so check the current terms on lmstudio.ai. Models you download are covered by their own licenses.

How do I enable GPU acceleration in LM Studio?

On NVIDIA: ensure CUDA drivers are installed. On AMD: ROCm is required. On Apple Silicon: Metal is used automatically. Go to Settings → GPU in LM Studio to verify GPU is detected and layers are offloaded.

What is the difference between Q4_K_M and Q5_K_M in LM Studio?

Q4_K_M uses 4-bit quantization (~4.5GB for 7B) with a small, model-dependent quality loss. Q5_K_M uses 5-bit (~5.7GB) with minimal loss. Use Q4_K_M for 8GB RAM; Q5_K_M or Q6_K for 16GB RAM systems.

Next Steps After Installing LM Studio

With LM Studio running, try Run Your First Local LLM to understand what response quality and speed to expect. For model recommendations matched to your hardware, see Best Beginner Local LLM Models. If you want to troubleshoot setup issues, see Troubleshooting Local LLM Setup.

Sources

  • LM Studio Official Website -- Downloads and documentation
  • Hugging Face Model Hub -- Full range of GGUF-quantized models
  • LM Studio on GitHub -- SDKs, CLI and bug tracker (the desktop app's own source code is not published there)

Common Mistakes When Installing LM Studio

  • Not allocating enough system RAM for the model you selected in LM Studio settings.
  • Using a pre-quantized model that is still too large for your GPU VRAM.
  • Expecting instant responses from large models on CPU-only systems -- response time will be 10-30 seconds.

A Note on Third-Party Facts

This article references third-party AI models, benchmarks, prices, and licenses. The AI landscape changes rapidly. Benchmark scores, license terms, model names, and API prices can shift between the time of writing and the time you read this. Before making deployment or compliance decisions based on this article, verify current figures on each provider’s official source: Hugging Face model cards for licenses and benchmarks, provider websites for API pricing, and EUR-Lex for current GDPR and EU AI Act text.

Run PromptQuorum with a local LLM, your own API keys, or both — you pick the backend.

Download the PromptQuorum Beta →

← Back to Local LLMs

MCP