LM Studio
Makes LM Studio, a desktop app for running open models offline.
- Type
- AI infrastructure
- Last shipped
- Nov 2, 2026
Products
Everything LM Studio shipped
November 2026
- New feature
LM Studio's Bionic now has a built-in terminal, toggled with Cmd+J or Ctrl+J, on Mac, Windows and Linux.
What this means for you: Bionic, LM Studio's app, now has a terminal built in, so you can type commands without leaving it. Press Cmd+J on Mac or Ctrl+J on Windows and Linux to show or hide it.
How to use it: Open Bionic on Mac, Windows or Linux. Press Cmd+J (Mac) or Ctrl+J (Windows, Linux) to toggle the terminal.
LM Studio's Bionic app now has a built-in terminal that I can open with Cmd+J or Ctrl+J. I am new to the command line. Explain what a terminal is and give me five safe, simple commands to try first, with what each one does. My computer is [Mac, Windows or Linux].
October 2026
- New feature
LM Studio's Bionic added Read Aloud, which reads text aloud using a small local text-to-speech model, first on Mac in version 1.1.9.
What this means for you: Bionic, the LM Studio app, can now read text out loud to you. It uses a small text-to-speech model that runs on your own computer, so the reading happens locally.
How to use it: Read Aloud is available first on Mac in Bionic version 1.1.9. The voice comes from Phonon, a model by Gradium.
- New model
LM Studio's Bionic added Reflection's Beam, a 501B mixture-of-experts model, hosted in the US with zero data retention.
What this means for you: You can now pick Beam, a very large 501-billion-parameter model from Reflection, inside Bionic, LM Studio's agent app. It runs on servers in the US, and zero data retention (ZDR) means the host does not keep what you send it.
Why it matters: The Oct 15 entry for Nemotron 3.5 Super VL said that model needs about 70GB of RAM to run locally. Beam is hosted in the US instead.
How to use it: Beam is live in Bionic now. Its open weights, under the Apache 2.0 license, are on Hugging Face at huggingface.co/reflection.
I use LM Studio's Bionic and it now offers Beam, a US-hosted model with zero data retention. Help me decide which of my tasks suit a hosted, no-retention model and which I should keep on my own computer. Ask me questions first. My typical tasks are: [list tasks].
- New feature
LM Studio's Bionic 1.1.8 adds support for the /v1/systemone and /v1/decisions APIs and can run at the same time as LM Studio.
- New feature
LM Studio's Bionic can now show live updating previews of documents, spreadsheets and slides.
September 2026
- New feature
LM Studio's Bionic now has a built-in interactive canvas for Excalidraw diagrams that both the user and Bionic can edit.
- New feature
LM Studio's Bionic agent can now search its own session history and reference other sessions in the composer with an @ mention.
- New feature
LM Studio's Bionic added local realtime voice transcription for Mac, Windows and Linux, running on Apple Silicon or NVIDIA GPUs.
- New model
LM Studio's Bionic added DeepSeek-V4.1-Flash, a model with native visual understanding.
August 2026
- Availability
LM Studio's Bionic is now available for Linux, running agents with local and open models.
- New feature
LM Studio's Bionic added Auto Review, which automatically approves shell tool requests using command parsing, matching and a reviewer subagent.
- New model
LM Studio's Bionic added GLM-5.3-Flash, which supports image input and is served from US servers with zero data retention.
Every LM Studio update is dated and explained on its own page. Read the full LM Studio explainer