LM Studio: Run AI Models Locally on Your PC

What Is LM Studio?

LM Studio is a desktop application-available for macOS, Windows, and Linux-that downloads GGUF-format models from Hugging Face, handles configuration through a GUI, and runs a local API server compatible with the OpenAI SDK. It is a desktop app that hides the awkward parts of local inference (finding a model, picking a quant that fits, wiring up an API) behind a UI that does not assume you have read a llama.cpp changelog.

The thing that separates it from the rest is that it ships two inference engines. llama.cpp runs GGUF files on NVIDIA, AMD, Intel and plain CPU. Apple's MLX runs MLX-format models natively on M-series Macs, and on the same chip an MLX build is commonly 10-40% faster than the equivalent GGUF. LM Studio picks the right one and lets you switch. The core app is proprietary, free for personal and commercial use as of late 2025.

Installation and Setup

Download the installer from lmstudio.ai. No package manager required, no dependency resolution. On macOS it's a standard DMG. On Windows, a standard installer. The Linux version (AppImage) has been in active development since late 2024 and is stable enough for daily use.

The first-launch experience streamlines model discovery:

• You land in a model browser that queries Hugging Face in real time. Search for "Qwen3 8B," click download, and it handles the rest-including checksum verification.
• Each listing shows a compatibility check against your machine before you download anything. LM Studio flags whether a given model will fit in your RAM or VRAM, so you are not left guessing why a 70-billion-parameter model refuses to load on a laptop.
• There is no separate step to convert formats or edit a config file.

One installer, a Hugging Face browser inside the app, and an OpenAI-compatible endpoint on port 1234 that any SDK will accept.

Supported Models

LM Studio supports a wide range of open-weight models through its Hugging Face integration. Common choices include:

• Qwen 3 (8B to 32B sizes, optimized for instruction-following)
• Llama series (from 3B to 70B parameter variants)
• Gemma 4 (instruction-tuned models with light hardware footprint)
• Mistral and Mixtral (efficient mid-range options)
• DeepSeek R1 and other reasoning-focused models
• Stable Diffusion XL-the open-weight image generator most commonly run inside LM Studio's diffusion mode.

Each model appears in multiple quantizations (Q4_K_M, Q5, Q8, FP16), allowing trade-offs between quality and memory usage.

Pros

• No command-line required: No command line is required. Everything is point-and-click.
• Cross-platform: It runs on Mac, Windows, and Linux, and it supports both GGUF and MLX model formats.
• Optimized for Apple Silicon: LM Studio 0.4.13 is the best GUI option for local LLMs right now, especially on Apple Silicon where its MLX backend significantly outpaces Ollama's GGUF path.
• Fast setup: It's become one of the most popular ways to run local models because it removes nearly all the friction. You install it, search for a model, click download, and start chatting.
• Extensible with plugins: The lm studio plugins system shipped in late 2024 and turned the application from a polished chat client into an extensible runtime. Plugins are written in TypeScript or Python, run inside the application's sandboxed worker, and can intercept inference requests, add new prompt processors, attach tool-calling backends, or expose entirely new interfaces. Common community plugins cover web search, code interpretation, retrieval-augmented generation over a local document folder, and integration with external API services that you choose explicitly.
• Privacy: LM Studio can operate entirely offline, just make sure to get some model files first.
• Document chat (RAG): You can attach documents to your chat messages and interact with them entirely offline, also known as "RAG".

Cons

• Closed source: The catch: it's not open source. It falls short for server deployments, scripted automation, and any context where you need auditable source code.
• Hardware demanding: Large models require significant GPU VRAM or system RAM, limiting accessibility on budget machines.
• Single-machine focus: It falls short for server deployments, scripted automation, and any context where you need auditable source code.
• Linux still in active development: While stable, the Linux AppImage receives less polish than Mac and Windows versions.

Best Use Cases

• Privacy-conscious work: Processing sensitive documents or prompts without cloud infrastructure.
• Experimentation and prototyping: Testing open-source models before deciding on cloud-based solutions.
• Development with local APIs: Building applications that call an OpenAI-compatible local endpoint.
• Code and research assistance: Running specialized models for coding, mathematics, or domain-specific tasks on your own machine.
• Offline operation: Environments with poor internet or security requirements preventing cloud usage.
• Educational exploration: Learning how LLMs work without paid API subscriptions.

FAQ

What hardware do I need to run LM Studio?

A 7B model at Q4 quantization needs approximately 4–5 GB, so 8 GB of total system RAM is a functional minimum. 16 GB is strongly recommended for comfortable multitasking alongside a browser and other tools. LM Studio runs on CPU-only hardware. A dedicated GPU with CUDA, ROCm, or Metal support speeds up inference considerably, but it is not required. On a modern quad-core CPU, a 7B Q4 model produces usable results-just at a slower token rate than GPU-accelerated setups.

Is LM Studio free?

The core app is proprietary, free for personal and commercial use as of late 2025.

Can I run LM Studio without a GUI, for example on a server?

llmster is the headless version of LM Studio, no desktop app required. It's ideal for servers, CI environments, or any machine where you don't need a GUI.

How do I integrate LM Studio with my own applications?

LM Studio provides a REST API that you can use to interact with your local models from your own apps and scripts. Point any OpenAI SDK at http://localhost:1234/v1 and it works with no code changes.