Short answer
- Ollama if you are happy typing one command, or want other apps (coding tools, agents) to use your model.
- LM Studio if you want a normal app window with a model browser, especially on a Mac.
- Jan if you want a chat window and the app itself must be open source.
What they have in common
All three download open models and run them on your own machine. Nothing is sent to a server. All three can act as a local “OpenAI-style” server, so other apps can talk to your model. Under the hood they mostly use the same engines (llama.cpp, and Apple’s MLX on Macs). Speed is close for the same model file, but can differ between apps and versions — check with your own model.
Ollama — the default
Ollama is a small background app with a command line. You pull a model by name and it runs. It is the easiest way to plug a local model into coding tools and agents. It is open source (MIT licence) and updates often.
ollama run qwen3.8:27bOn Macs, Ollama also offers MLX builds of some models, with names ending in -mlx (for example qwen3.8:27b-mlx). They can be faster on Apple Silicon. The plain name downloads the regular build. Each model page on this site lists the exact name to use.
LM Studio — the friendly window
LM Studio is a desktop app: search for a model, click download, chat. It shows which versions fit your memory before you download them, and on a Mac it offers MLX versions. It is free for local use, but the app itself is closed source.
Since 16 July 2026, the main download on lmstudio.ai is LM Studio Bionic, a separate agent app with optional paid cloud plans. The classic LM Studio app is still available.
Jan — the open-source window
Jan looks like a chat app and runs fully offline by default. Its code is open under the Apache 2.0 licence (we read the licence file on 10 October 2026). Version 0.8.6 came out on 9 October 2026. Its model browser is simpler than LM Studio’s.
Which one for…
| You want… | Pick |
|---|---|
| The quickest start from a terminal | Ollama |
| A coding assistant or agent using your model | Ollama |
| To click, not type, on a Mac | LM Studio |
| To see what fits before downloading | LM Studio (or our checker) |
| Open-source software only | Jan or Ollama |
| An AMD graphics card on Windows or Linux | Read our AMD guide first |
What about llama.cpp itself?
llama.cpp is the engine inside most of these apps. You can use it directly (llama-server) for the most control over settings. It updates almost every day, so expect to update often. Most people do not need it.