If you’re new to LM Studio, I’ve written an Introduction to LM Studio — that post covers the UI, downloading models, chat, parameters, and using it as a server. I gave a talk about this, and someone suggested I look into Docker Desktop — apparently, Docker Model Runner does the same thing.

I created this post so that I could investigate and see how they compare.

What They Have in Common

Both approaches aim to let you:

• Run an LLM locally.
• Expose an OpenAI-compatible REST API so existing code and tools can call chat/completions (and related endpoints); that is, you can use them for prototyping.
• Download models and then then reuse them across sessions.

You can switch between the two easily by simply changing the port.

LM Studio

LM Studio is a desktop application; it provides a catalogue, chat, parameters, Developer tab, optional MCP. The walkthrough (with screenshots) is in Introduction to LM Studio; here’s a brief summary.

Typical workflow:

1. Install LM Studio
2. Download a model (e.g. Gemma, Llama, Qwen) from the built-in catalogue
3. Load the model and start the server on the Developer tab (default API base: http://localhost:1234/v1/)
4. Call it code or another client

Strengths

• The UI is purpose built for exactly this
• It’s a familiar workflow / interface
• Provides MCP Integration
• You don’t need docker to run

Down-sides

• Dedicated app on the machine
• Lm Studio seems to have a separate section of Hugging Face

Docker Desktop and Model Runner

Docker Desktop now has a feature called Docker Model Runner. It’s still Beta, but adds a local LLM inference. You enable it in settings, pull models from Docker Hub (or Hugging Face), and get an OpenAI-compatible endpoint (http://localhost:12434/engines/v1) on the host (note the different port and path from LM Studio). You would be forgiven for thinking that one of these products have “taken inspiration” from the other!

Typical workflow:

1. Install or update Docker Desktop
2. Enable Model Runner (Settings → AI, or see Get started with Docker Model Runner)
3. Pull a model, e.g. docker model pull (see Docker Model Runner and the API reference)
4. Call http://localhost:12434/engines/v1/chat/completions from your app

Strengths

• It’s one less dedicated app, if Docker Desktop is already there
• It’s the same idea as docker: pull, version, script
• Has the same OpenAI-compatible API as LM Studio

Trade-offs

• Still in Beta
• It’s a different default URL - you’ll need to update BaseUrl in config (1234 vs 12434, and the /engines/v1 path)
• If you don’t already use Docker Desktop then this probably isn’t going to work for you - it’s a heavy app for just this!
• The UI is (obviously) less focused on the LLM part - since it’s first and foremost a container management app

Some Code

From .NET, Python, etc.., the pattern is the same — only the URL changes. LM Studio:

{
  "LmStudio": {
    "BaseUrl": "http://localhost:1234/v1/",
    "Model": "google/gemma-3-4b"
  }
}

Docker Model Runner:

{
  "LocalLlm": {
    "BaseUrl": "http://localhost:12434/engines/v1/",
    "Model": "your-model-id-from-docker-model-runner"
  }
}

Which Wins?

To be honest, I’m probably going to switch to Docker Desktop (if I notice any quirks then maybe I’ll go back). Since I run Docker anyway, it means I can bin LM Studio. However, if you don’t already use Docker then LM Studio is probably a little easier to use.

References

DDD South West

DDD Conferences (UK)

Introduction to LM Studio

LM Studio

Docker Model Runner

Get started with Docker Model Runner

Docker Model Runner API reference

Run LLMs Locally with Docker (blog)

Programmatic Interaction with a Local LLM in .NET