Two popular ways to run a local chat model on a Windows PC are LM Studio, a ready-made desktop app, and Text Generation WebUI, the open-source project started by oobabooga. Both can load models on your own hardware and keep the conversation on the machine. They are aimed at slightly different people.
This comparison uses each project's official documentation and sites, read for this article, and it lines up with the way we already describe LM Studio in Jan vs LM Studio and Ollama vs LM Studio.
What each one is
LM Studio is a free desktop application from the official LM Studio site. You install it, browse a catalogue of local models, chat in a desktop window, and, if you want, turn on a built-in OpenAI-compatible local server so other apps can talk to the same model. We walk through that server in How to use LM Studio's local server and through first-time setup in Ollama or LM Studio. It is the polished, guided option, and it is not an open-source app in the same sense as Text Generation WebUI.
Text Generation WebUI (often called oobabooga, after its original author) is an open-source local workbench. Officially you get it from the project's GitHub repository, and on Windows the usual path is the project's official start script or a portable build published with the release. It then serves a browser UI on the local machine, commonly at http://127.0.0.1:7860. The point of the project is flexibility: multiple model back-ends, a lot of extensions, and fine-grained settings that power users expect to see. That flexibility is also why a first launch asks more of you than LM Studio does.
If you only want a background service and a simple command line, Ollama remains the third option, including for larger models such as the ones in our local-AI news. This page is only about the two all-in-one front ends.
How they differ
Topic | Text Generation WebUI | LM Studio |
|---|
Licence and code | Open-source project you can inspect on GitHub | Free desktop app, distributed as a ready-made build |
Interface | Browser-based web UI | Native desktop app with a model catalogue and chat window |
Setup | Official install script or portable zip; you pick GPU back-ends and options | Installer and a guided catalogue |
Extensions and loaders | A major reason to use it: many back-ends and extensions | Fewer moving parts; less tinkering |
Local API | Can be enabled as part of the project | First-class, documented OpenAI-compatible server |
Typical user | People who want maximum control and an inspectable stack | People who want a clear app and a server for other software |
There is no official "which is smarter" score. Output quality depends on the model you load, the quantisation, and your GPU or RAM, not on which window you typed into. If a model runs in both, treat any difference you notice as something to verify, not as a reason we can declare a winner.
When to choose LM Studio
Choose LM Studio if you are new to local models, or if you want a catalogue, a chat window and a documented local server without editing config files. That is why it sits in our beginner guide, in how to choose models in LM Studio, and in the best free local AI apps list. It is also the easier pair with tools that already speak an OpenAI-style API.
When to choose Text Generation WebUI
Choose Text Generation WebUI if you want the open-source workbench: extra loaders, extensions, and settings that a simple desktop app may not expose. It is a better second or third install than a first one, unless you already like running Python-style local tools. Download it only from the official GitHub project, and read that repository's current Windows instructions, because installers and default ports can change between versions.
If your real goal is document chat rather than a flexible text-generation workbench, our AnythingLLM guide or Open WebUI versus AnythingLLM may be a closer match than either of these two apps.
Our verdict
Start with LM Studio if you want the straightforward local chat app and a documented OpenAI-compatible server. Choose Text Generation WebUI if you want an open-source, highly configurable workbench and you are happy to follow the official GitHub install. Both run models on your PC; neither is "smarter" in a way we can back with a test.