LM Studio
Free desktop app for running local LLMs with llama.cpp and MLX, OpenAI- and Anthropic-compatible local server, MCP host and headless llmster daemon
Overview
LM Studio is a desktop app from Element Labs for downloading and running open LLMs on your own computer, with llama.cpp and Apple MLX engines under the hood. You browse and download models, chat with them, and serve them through a local server that speaks OpenAI-style endpoints (chat, responses, embeddings, structured output, tool use) and, since version 0.4.1 in January 2026, an Anthropic-compatible /v1/messages endpoint, so tools like Claude Code can run on local models. It has been an MCP host since version 0.3.17 (June 2025). For servers and CI it ships llmster, the LM Studio core as a headless daemon, plus the open-source lms CLI and Python and TypeScript SDKs (MIT). Since July 2025 the app is free for both home and work use, with no commercial license form. The desktop app itself is closed source. In July 2026 the same team launched Bionic, a separate agent app for coding and document work on open models that can also call cloud-hosted open models through LM Studio's Zero Data Retention cloud. Also from LM Studio: LM Studio Bionic, its agent app for open models.
The Verdict
Who Should Use LM Studio?
Best For
- Developers who want a GUI to find, test and compare local models
- Mac users who want MLX speed on Apple Silicon
- Pointing OpenAI or Anthropic clients at a local model
Not Ideal For
- Policies that require open-source, auditable code
- Multi-user serving with many parallel requests
What's Great
- Easy GUI: install, pick a model, start chatting
- MLX by default on Mac; faster than Ollama in one M4 Pro test
- Built-in MCP with tool-call confirmation
Watch Out For
- Desktop app is proprietary; only the CLI and SDKs are open source
- Server stops when the app closes unless you run llmster
- Requests are queued, which limits multi-user serving
Pricing
View all features & details
Key Features
- Model search and download inside the app
- llama.cpp and MLX engines
- OpenAI-compatible and Anthropic-compatible local endpoints
- REST API for stateful chats, model load, unload and download
- MCP host, including MCP servers over HTTP
- llmster headless daemon, lms CLI, Python and TypeScript SDKs
- LM Link for using your local models from other devices
Company
- Element Labs, founded by Yagil Burowski
- Office in the Brooklyn Navy Yard, per job posts
- Also makes Bionic, an agent app for open models (Jul 2026)
Real-World Usage
Developer Use
- Official guide for running Claude Code on local models
- Frequent comparisons with Ollama and llama.cpp in 2026 reviews
Community
- MIT lms CLI and SDKs, CLI pushed Nov 10, 2026
- Free public Hub organizations for sharing presets
How It Compares
| Feature | LM Studio | Ollama |
|---|---|---|
| Interface | Full desktop GUI | CLI and background service |
| License | Proprietary app, MIT CLI and SDKs | MIT |
| Mac engine | MLX by default | Moving to MLX |
| Headless server | llmster daemon | Native background service |
| Parallel requests | Queued | Handled in parallel |
Sources: LM Studio · Download (0.4.28) · Developer docs · MCP (Jun 2025) · Free for work (Jul 2025) · Bionic (Jul 2026) · Morph comparison · see also Ollama and llama.cpp
Is this your tool? Add a badge