LM Studio vs Ollama: license, API, engines and cloud plans compared

Updated · 7 min read

LM Studio and Ollama both download open models, run them on your own computer and serve them to other apps through an OpenAI-compatible API, and both are free for local use. Ollama is open source (MIT), works mainly from the terminal, runs as a background server and is the runtime most apps support. LM Studio is a closed-source but free desktop app with a full interface for searching, tuning and chatting with models, plus an optional headless server. If you want a GUI on a Mac or Windows PC, start with LM Studio. If you want open source, a Linux server or the widest app support, use Ollama. Facts below were checked on each project's docs and download pages on 2026-11-18.

Side by side

Ollama LM Studio
License MIT, open source Proprietary, free for personal and internal business use; the lms CLI, SDKs and MLX engine are MIT
Latest version 0.40.16 (2026-11-09) 0.4.28 (2026-10-31)
Interface CLI, plus a desktop app with chat on macOS and Windows Desktop app, plus the lms CLI
Headless server ollama serve, systemd service on Linux, official Docker image llmster daemon for Linux, macOS and Windows
Default address http://localhost:11434 http://localhost:1234
APIs Native /api, OpenAI-compatible /v1 (chat completions, responses, embeddings), Anthropic Messages Native REST /api/v1, OpenAI-compatible /v1 (chat completions, responses, embeddings), Anthropic /v1/messages
Keys on the local API None: the local API needs no authentication Optional API tokens with permissions
Engines llama.cpp for GGUF; MLX on Apple Silicon, the default for supported models since 0.40.0 llama.cpp for GGUF; MLX on Apple Silicon
Where models come from Ollama's library; import GGUF or Safetensors with a Modelfile Hugging Face search inside the app; lms import for a GGUF file (experimental)
Platforms macOS 14+ (Apple Silicon; Intel Macs CPU only), Windows 10 22H2+, Linux, Docker macOS 14+ on Apple Silicon only, Windows x64 and ARM, Linux x64 and ARM64
Cloud models Yes, in the same tool: pay per token, or Pro $20, Max $100, Team $500 a month Only in the separate Bionic app: Bionic+ $20 or Pro $100 a month
Apps listed on BYOKList 46 21, all of which also support Ollama

Ollama

Ollama installs a background server and a command-line tool. ollama run gemma4:e2b downloads a model and starts a chat in the terminal; on macOS and Windows the desktop app adds a chat window with file drag-and-drop. Since version 0.40.0 (September 2026), models whose architecture the MLX runtime supports run on MLX automatically on Apple Silicon. Everything else runs on llama.cpp with GGUF files.

The server speaks three APIs on one port: Ollama's own at /api, an OpenAI-compatible one at /v1 and an Anthropic-compatible Messages endpoint, which is what lets Claude Code run on local models. ollama launch claude, ollama launch codex and ollama launch opencode set those agents up for you.

Ollama is also a cloud service. After ollama signin, model names ending in :cloud run on Ollama's servers through the same local API. Cloud use is not free: you pay per token, or take Pro ($20 a month or $200 a year, $60 of usage included), Max ($100, $300 included, 10 concurrent requests) or Team ($500, in early access, unlimited users, $1,000 included). Ollama says it does not store, log or train on cloud prompts. To stay fully local, set OLLAMA_NO_CLOUD=1.

Three defaults catch people out. The server binds to 127.0.0.1, so set OLLAMA_HOST=0.0.0.0:11434 to reach it from other devices. The default context window is 4,096 tokens, small for coding agents; raise it with OLLAMA_CONTEXT_LENGTH. The local API has no authentication, so anyone who can reach the port can use it; the docs show how to put it behind a proxy such as Nginx.

LM Studio

LM Studio is built around its desktop app. You search and download models from Hugging Face inside it, chat with them, attach documents for offline RAG, connect MCP servers and switch between the llama.cpp and MLX runtimes (Cmd/Ctrl+Shift+R).

The server starts from the Developer tab or with lms server start. It offers LM Studio's own REST API (/api/v1, with stateful chats, MCP and model download, load and unload), OpenAI-compatible endpoints including /v1/responses, which Codex needs, and an Anthropic-compatible /v1/messages for Claude Code. Server settings include API tokens with permissions, a Serve on Local Network switch, CORS and just-in-time loading, which loads a model when a request names it. Since 0.4.0 (January 2026), parallel requests to one model use continuous batching on the llama.cpp engine. For machines without a screen, llmster packages the LM Studio core as a daemon (lms daemon up), and LM Link lets one device use a model loaded on another (up to 5 devices on the free plan).

The app is closed source. Since July 8, 2025 it has been free for use at work without a separate commercial license. The current terms (version of August 23, 2026) allow use "for Your personal and / or internal business purposes" and forbid modifying or reverse engineering it, redistributing it, or running it as a service bureau, application service provider or software-as-a-service. You can run it for your own team, but not resell it or host it for customers; Ollama's MIT license has no such limits.

LM Studio's cloud models live in Bionic, a separate agent app launched in July 2026. Bionic is free with local models. Bionic+ ($20 a month) adds US-hosted open models such as Kimi K3, GLM 5.3 and DeepSeek V4 Flash, and Pro ($100) raises the limits and includes usage of GPT Sol 6.1 and Claude Opus 5.5. LM Studio says its cloud keeps no prompts or responses (zero data retention).

Connecting either one to a BYOK app

Setting Ollama LM Studio
OpenAI-compatible base URL http://localhost:11434/v1 http://localhost:1234/v1
Anthropic base URL (Claude Code) http://localhost:11434 http://localhost:1234
API key field Any text, such as ollama (ignored) Leave empty or type anything; use an LM Studio token if Require Authentication is on
Browser apps and extensions Allow the origin with OLLAMA_ORIGINS Turn on Enable CORS in Server Settings
Model names Ollama tags such as gemma4:e2b The identifier LM Studio shows, such as openai/gpt-oss-20b

Many apps have a dedicated Ollama option that uses the native API at http://localhost:11434. With LM Studio, turn on just-in-time loading if you want the app's model list to show every downloaded model, not only the ones loaded in memory.

Which should you pick

Your situation Pick Why
You want a desktop app to browse, test and tune models LM Studio Hugging Face search, runtime switching and load settings are all in the GUI
You need open source, or want to ship the runtime inside a product Ollama MIT license; LM Studio's terms forbid redistribution and hosted-service use
Linux server, Docker or a GPU box Ollama, or LM Studio's llmster Ollama has an official Docker image and a systemd service; llmster runs without the GUI
One server shared on your network LM Studio, or Ollama behind a proxy LM Studio has API tokens built in; Ollama's local API has no authentication
Intel Mac Ollama LM Studio does not support Intel Macs; Ollama runs there on the CPU
The app you want lists only one That one 46 apps on BYOKList support Ollama and 21 support LM Studio
Models bigger than your computer can hold Ollama cloud or LM Studio Bionic Both charge per token or monthly; compare what each plan includes

You can install both: they use different ports. Each keeps its own model folder (Ollama uses ~/.ollama/models on macOS), so a model downloaded in LM Studio has to be pulled again in Ollama or imported with a Modelfile.

Jan and AnythingLLM as a third option

Jan is the closest open-source match for LM Studio: an Apache 2.0 desktop app for macOS, Windows and Linux that downloads models from Hugging Face, runs them on llama.cpp or, on Apple Silicon, MLX, and starts an OpenAI-compatible server at http://127.0.0.1:1337/v1. It also takes your API keys for OpenAI, Anthropic, Mistral, Groq and others, and can use Ollama as a custom endpoint. Pick Jan over LM Studio if open source matters to you. Pick LM Studio if you rely on MLX models (Jan marks its MLX engine experimental, without embeddings) or want LM Link and a headless daemon. Jan's bundled llama.cpp does not load models on Intel Macs as of 0.8.5.

AnythingLLM covers a different need: chatting with your documents and running agents. It is MIT-licensed and comes as a single-user desktop app (macOS, Windows, Linux) or a multi-user Docker server, and it can use Ollama or LM Studio as its model provider alongside cloud keys. In October 2026 its built-in engine moved from a bundled copy of Ollama to llama.cpp managed by AMD's Lemonade; connecting your own Ollama still works. So AnythingLLM vs LM Studio is rarely a choice between the two: run models in LM Studio and use AnythingLLM as the workspace on top.

For other runtimes, including llama.cpp's own server, LocalAI and vLLM, see Ollama alternatives. For chat apps and coding tools that connect to both, see local LLM apps.

Apps that work with Ollama or LM Studio

  • Jan

    Free, open source · macOS, Windows, Linux · Open source

    Visit

    Free, open-source ChatGPT alternative that runs models on your computer or connects to online models with your API key.

    Works with:OpenAIAnthropicGoogle GeminiMistral+2 more

  • AnythingLLM

    Free · macOS, Windows, Linux · Open source

    Visit

    Private AI app for chatting with your documents. Runs local models or connects to OpenAI, Anthropic, Gemini and others with your key.

    Works with:OpenAIAzure OpenAIAnthropicGoogle Gemini+9 more

  • Infinity

    Free · macOS

    Visit

    Free Mac AI chat app with git-style branching: fork any part of an answer into its own thread, ask up to four models at once, and use your own keys or local models.

    Works with:OpenAIAnthropicGoogle GeminixAI+7 more

  • Caotang

    Free · Web · Open source

    Visit

    Free, local-first browser writing studio for novels and scripts, with an AI agent that runs on your own Anthropic, OpenAI, Google, OpenRouter or local model key.

    Works with:AnthropicOpenAIGoogle GeminixAI+5 more

  • Laya

    Free · macOS, Windows, Linux · Open source

    Visit

    Open-source, local-first desktop command center that turns Slack, Gmail, GitHub, Jira and calendar notifications into AI-drafted action cards, using local models or your own API keys.

    Works with:AnthropicOpenAIGoogle GeminiOllama+1 more

  • MergeJudge

    Free AI reviews on public repos; bring your own key for private repos · Web, Self-hosted · Open source

    Visit

    AI pull request reviewer offered as a hosted GitHub App or an open-source GitHub Action, running on your own Anthropic, OpenAI or OpenAI-compatible key, local Ollama, or rules only.

    Works with:AnthropicOpenAIOllama

  • Conduit

    Free · Windows, macOS, Linux · Open source

    Visit

    Open-source desktop AI app for agents, research, writing and slides that runs on your own keys from 23 providers or local models.

    Works with:AnthropicOpenAIGoogle GeminixAI+12 more

  • SoloMD

    Free · macOS, Windows, Linux · Open source

    Visit

    Free, open-source Markdown editor for desktop and mobile with built-in AI tools that run on your own key from 14 providers.

    Works with:OpenAIAnthropicGoogle GeminixAI+7 more

See all 47 apps that work with Ollama

Questions

Is LM Studio better than Ollama?

Neither wins across the board. LM Studio has the fuller desktop app for finding, tuning and chatting with models. Ollama is open source, is easier to run as a server on Linux or in Docker, and more apps support it. Both are free for local use.

Is LM Studio open source?

No. The LM Studio app is proprietary, though free for personal and internal business use. Its lms CLI, SDKs and MLX engine are MIT-licensed on GitHub. Ollama is fully open source under the MIT license.

Can I use LM Studio at work?

Yes. Since July 8, 2025 LM Studio needs no separate commercial license for work. Its terms allow personal and internal business use, but not reselling it or running it as a hosted service for others.

Which is faster, LM Studio or Ollama?

Both run GGUF models on llama.cpp and use Apple's MLX on Apple Silicon Macs, so the engine underneath is often the same. Speed on the same model and quantization mostly comes down to settings such as context length and GPU offload, so test the model you plan to use on your own machine.

What ports do Ollama and LM Studio use?

Ollama listens on http://localhost:11434, with the OpenAI-compatible API at /v1. LM Studio's server uses port 1234 by default, so its OpenAI-compatible base URL is http://localhost:1234/v1.

Is Ollama free?

Running models on your own computer is free and unlimited. Ollama's cloud models cost money: pay per token, Pro at $20 a month with $60 of usage included, or Max at $100 a month with $300 included.

Sources

Checked against the providers' own documentation on 2026-11-18. Prices and limits change; the links below have the current figures.

Related guides