LM Studio
Desktop app for running open-weight models locally, with a GGUF and MLX runtime, an OpenAI-compatible local server, a CLI and Python and TypeScript SDKs.
Updated 2026-08-02
Not settled here. This entry does not record a confirmed free tier for LM Studio.
Listed pricing: Free desktop app; cloud inference is pay as you go per token.
Overview
LM Studio is the graphical route to running open-weight models on your own computer. It ships for macOS, Windows and Linux, covering Apple Silicon Macs, x64 and ARM64 Windows PCs and x64 Linux boxes, and it runs both llama.cpp GGUF models and, on Apple Silicon, Apple's MLX format. The documentation names Llama, DeepSeek, Qwen, Phi, Mistral and gpt-oss among the families it runs.
The part developers actually care about is the server. LM Studio exposes an OpenAI Compatibility API plus its own REST API, still marked beta, so anything already written against the ChatGPT API format can point at localhost and keep working. Around that sit a headless CLI called lms that handles chat, model downloads, daemon management, server control and publishing, and first-party SDKs for Python (lmstudio-python) and TypeScript (lmstudio-js).
My read: the product is currently two products, and the site does not fully reconcile them. The documentation describes the mature local-model desktop app above. The front page now leads with LM Studio Bionic, "An Agent made for Open Models", described as "Natively local" and built "for creativity, work, and code", with local real-time voice transcription and agentic control of coding tasks and automations. Bionic is in initial preview for macOS and Windows only, its runtime uses MLX and llama.cpp underneath, and it can also reach frontier open models the page names as GLM 5.2, Kimi K3 and DeepSeek V4 Pro. Pricing reflects that split: the local app is free, cloud inference is pay as you go with per-token rates, and the Bionic Pass tier says only "Pricing and plan details coming soon". The pricing page also says nothing about commercial or business licensing terms, which is worth resolving directly with the vendor before rolling it out at work.
Is LM Studio free?
Not settled here. This entry does not record a confirmed free tier for LM Studio.
What the entry records: Yes. The desktop app and local inference are free. The pricing page does not state commercial or business licensing terms either way, so a team rolling it out at work should confirm those with the vendor.
Listed pricing: Free desktop app; cloud inference is pay as you go per token.
No figure is estimated here for what is not published. Check LM Studio's own site for the current terms.
Pricing verified . That is the date the plans were last re-checked against the vendor's own pages, not today's date. Confirm on the official site before you pay.
LM Studio pricing
| Plan | Price | What's included |
|---|---|---|
| Free | $0 | Local LLMs and voice transcription on your own machine. LM Studio's wording: "No data ever leaves your device." |
| Pay as you go | $0.95 to $3.00 per million input tokens, $3.48 to $15.00 per million output tokens | Cloud inference credits. The rate depends on which model you pick, so the spread is wide by design. |
| Bionic Pass | Not published | The pricing page says only "Pricing and plan details coming soon" for the Bionic tier. |
Local LLMs and voice transcription on your own machine. LM Studio's wording: "No data ever leaves your device."
Cloud inference credits. The rate depends on which model you pick, so the spread is wide by design.
The pricing page says only "Pricing and plan details coming soon" for the Bionic tier.
One plan carries no published number: Bionic Pass, recorded as “Not published”. Nothing is estimated in its place. The vendor's own pricing page is the only source for those figures.
Pricing verified . That is the date the plans were last re-checked against the vendor's own pages, not today's date. Confirm on the official site before you pay.
Is LM Studio worth it?
Worth it for People who want a graphical way to find, download and run open-weight models locally, plus a drop-in OpenAI-compatible endpoint for their own code.
There is no free tier to test it on, so the plans above are the entry cost. The recorded trade-offs are listed below, and any one of them can settle the question on its own.
The 8.5/10 AI Score is an editorial read of published capability, price and shipping pace. Nobody here has hands-on hours with LM Studio. How we verify.
Worth it if
The strengths recorded against this entry.
- Graphical discovery, download and management of open-weight models, which is a much lower barrier than a terminal-only runtime
- Runs both GGUF and, on Apple Silicon, MLX models from the same application
- OpenAI-compatible local endpoint means existing code needs a base URL change and nothing else
- Headless lms CLI and first-party Python and TypeScript SDKs make it viable for automation, not just interactive chat
- Local use keeps data on the device, and the cloud services are stated as zero data retention
Not worth it if
Any one of these blocks your use case.
- The site currently presents two products at once, with docs describing the mature desktop app and the front page leading with the Bionic agent preview
- Bionic is preview-only and macOS and Windows only, with no published price for the Bionic Pass tier
- The pricing page says nothing about commercial or business licensing, so work use needs a direct answer from the vendor
- Local model quality is bounded by your hardware, and the cloud path reintroduces per-token cost
- Cloud token rates span a wide range depending on model, so a bill is hard to forecast from the pricing page alone
What sets LM Studio apart
- Runs both llama.cpp GGUF models and Apple MLX models from the same app
- Exposes an OpenAI Compatibility API locally, so existing code points at localhost with no rewrite
- Headless `lms` CLI covers chat, downloads, daemon management, server control and publishing
- First-party Python and TypeScript SDKs rather than community wrappers
- LM Studio Bionic adds an agent layer with local voice transcription, in preview on macOS and Windows
Key features
GGUF and MLX in one app
Runs llama.cpp GGUF models across platforms and Apple's MLX format on Apple Silicon, so the same interface covers both the portable route and the Mac-native one without swapping tools.
OpenAI-compatible local server
Serves models through an OpenAI Compatibility API, alongside an LM Studio REST API still marked beta. Existing code written for the ChatGPT API format points at localhost and keeps working.
lms command-line tool
A headless CLI covering chat, model downloads, daemon management, server control and publishing, which is what makes the app usable on a box with no desktop session.
Python and TypeScript SDKs
lmstudio-python handles model loading, text generation, embeddings and agents; lmstudio-js covers Node and TypeScript. First-party SDKs rather than community wrappers, which matters for anything long-lived.
LM Studio Bionic (preview)
An agent layer described as "Natively local", with document creation and editing, automatic saving, local real-time voice transcription in multiple languages, and agentic handling of coding tasks, automations and computer control. In initial preview for macOS and Windows.
Cloud access to larger open models
Beyond local inference, LM Studio offers cloud models including ones the site names as GLM 5.2, Kimi K3 and DeepSeek V4 Pro, under a policy it states as "Zero Data Retention (ZDR) across the board".
How it compares
| Tool | Best for | Pricing | Score |
|---|---|---|---|
| LM Studio | People who want a graphical way to find, download and run open-weight models locally, plus a drop-in OpenAI-compatible endpoint for their own code. | Free desktop app; cloud inference is pay as you go per token | 8.5/10 |
| Ollama vs Ollama → | Anyone who wants open-weight models running on their own hardware for privacy, offline work or zero marginal cost, with a cloud fallback for models that will not fit. | Free, with Pro at $20/mo and Team at $25/seat/mo for cloud usage | 8.8/10 |
| Osaurus | Developers on Apple Silicon Macs who want a private, local LLM server for coding assistants and automation without sending data to the cloud. | Free (open-source) | 7.5/10 |
| DeepSeek vs DeepSeek → | Developers and teams running heavy coding or reasoning workloads who want frontier-level model performance without frontier-level API bills. | Free chat + API from $0.27/M input tokens | 8.9/10 |
| GLM-5.2 vs GLM-5.2 → | Development teams building coding agents who want a frontier-class model they can self-host, fine-tune, and deploy under a fully permissive license. | Free 20M tokens on signup + open weights; paid API & coding plans (check site) | 8.7/10 |
Compare head-to-head
Related reading
Claude Eval Escapes: An 8-Day Disclosure Clock
Anthropic suspended the cyber evals July 23, notified affected organizations July 27, published July 31. What those dates settle and what they don't.
DeepSeek-V4-Flash API: Responses, Codex, Agent Setup
DeepSeek opened the V4-Flash public beta on July 31, 2026 with a native Responses API. Model ID, server-side state, Codex setup, and the agent claim.
Three Claude Eval Escapes in 141,000 Runs
Anthropic's July 30 disclosure came out of a retrospective sweep of 141,000+ cyber eval runs. The chronology, the arithmetic, and what it predicts.
Ready to try LM Studio?
Head to the official site to start with LM Studio — pricing and plans are listed above.
Visit LM Studio