SoftNexi

Windows · Download, Screen Capture & AI Tools

LM Studio for Windows

Desktop app for downloading and running open-weight large language models locally, with a chat interface and an OpenAI-compatible local API server.

Updated August 9, 2026 · Reviewed by SoftNexi Editorial Team, Software research and documentation

Download

Overview

LM Studio provides a graphical way to discover, download, and run open-weight large language models (such as those in the Llama, Mistral, or Qwen families) entirely on your own PC, without sending prompts to a cloud provider. It wraps local inference engines behind a chat interface similar to familiar AI chat apps, plus a built-in model browser for finding compatible model files.

Because inference runs locally, hardware matters a lot: larger models need more system memory and, for good performance, a GPU with sufficient video memory (VRAM) to hold the model — LM Studio also supports CPU-only inference, which works but is generally much slower. Smaller, more heavily quantized models run comfortably on modest hardware, while larger models need a capable GPU to feel responsive; check a specific model's size and quantization against your hardware before downloading it, rather than assuming any model will run well.

LM Studio also runs a local server that exposes an OpenAI-compatible API on your own machine, which lets developers point existing tools and scripts at a local model instead of a cloud API, without prompts ever leaving the device.

The core privacy benefit of local inference is straightforward: prompts and generated text stay on your computer rather than being sent to a third-party server, which matters for sensitive documents, proprietary code, or anyone who simply prefers not to share their queries with a cloud AI provider. LM Studio is free to download and use.

Key features

  • Built-in model browser for discovering and downloading open-weight LLMs
  • Chat interface for local conversation with a downloaded model
  • Local OpenAI-compatible API server for use with existing developer tools
  • GPU acceleration where supported hardware and drivers are present, with CPU-only fallback
  • Support for common quantized model formats to reduce memory requirements
  • No account or internet connection required to run a model once it's downloaded

System requirements

Operating system
Windows 11 and Windows 10 (64-bit)
GPU
A dedicated GPU with enough VRAM to hold your chosen model significantly improves speed; CPU-only inference is supported but slower
Disk space
Varies widely by model — individual model files can range from a few gigabytes to well over ten, so check a model's listed size before downloading

How to install LM Studio

  1. 1. Download the installer

    Get the Windows build from lmstudio.ai/download.

  2. 2. Run the installer

    Setup completes without bundled offers.

  3. 3. Browse and download a model

    Use the built-in model search to find an open-weight model sized appropriately for your GPU's VRAM or available system memory.

  4. 4. Load the model and chat

    Select the downloaded model from the chat tab and start a conversation to confirm it runs at an acceptable speed on your hardware.

How to use it

  1. 1. Match model size to your hardware

    Start with a smaller, more heavily quantized model if you're unsure whether your GPU has enough VRAM, then try larger models if performance is good.

  2. 2. Run the local API server

    Enable the local server feature to expose an OpenAI-compatible endpoint on your machine for use with other apps or scripts.

  3. 3. Use CPU-only mode if needed

    If you don't have a compatible GPU, LM Studio can still run models on the CPU, though generation will be noticeably slower.

  4. 4. Keep sensitive prompts local

    Use a locally run model instead of a cloud AI service when working with confidential documents or proprietary code you don't want leaving your device.

Safety and privacy

  • Download only from lmstudio.ai/download; the app itself is closed-source but runs openly published model weights you choose to download.
  • Model files come from third-party repositories (such as Hugging Face) linked within the app — only download models from sources you trust, since a model file is a large binary you're running on your machine.
  • We link only to the developer's own site and have not independently audited the current build.
Bundled software
The official installer does not bundle third-party offers.
Privacy
Once a model is downloaded, inference runs entirely on your device; prompts and generated output are not sent to a cloud service unless you specifically configure the app to connect to one.

Known risks

  • Running large models on hardware without enough VRAM or system memory can cause slow performance or the app running out of memory — check requirements before downloading a specific model.
  • Locally run open-weight models can still produce inaccurate or inappropriate output; local inference removes cloud data exposure, not model error.

What's new

    Pros and cons

    Pros

    • Prompts and data stay on your device, not sent to a cloud provider
    • No account or subscription required to use downloaded models
    • OpenAI-compatible local API for plugging into existing tools
    • Works without an internet connection once a model is downloaded

    Cons

    • Performance is heavily dependent on your GPU's VRAM and system memory
    • Larger, more capable models require substantially more hardware than smaller ones
    • Model quality and behavior vary by which open-weight model you choose, not by LM Studio itself
    • The app itself is closed-source, even though it runs open-weight models

    Verdict

    LM Studio is the most approachable way for Windows users to run open-weight LLMs locally without touching a command line, provided their hardware — particularly GPU VRAM — matches the model they want to run. It's a strong privacy-first alternative to cloud AI chat tools for anyone comfortable choosing and managing model files.

    Frequently asked questions

    Is LM Studio free?

    Yes, the app is free to download and use.

    Do I need a powerful GPU to use LM Studio?

    A GPU with sufficient VRAM makes inference much faster, but LM Studio also supports CPU-only inference, which works but is slower. Smaller quantized models run acceptably on more modest hardware.

    Is my data private with LM Studio?

    Once a model is downloaded, prompts and generated text stay on your device and are not sent to a cloud service unless you specifically configure that.

    Does LM Studio work on Windows 11?

    Yes, the official installer supports Windows 11 and Windows 10, including ARM64 devices.