LM Studio 0.4.28, released on 31 October 2026, fixes a Mac fault where choosing "Restart to apply update" reopened the old version, and Bionic 1.2.0, released on 5 November 2026, adds a new engine for Apple silicon Macs called oMLX. If LM Studio is working for you, there is no urgent reason to update. Update if your Mac keeps reopening the old version, if you have an M5 Mac or an AMD Radeon 8065S, or if you run the Flint coding model.
LM Studio is the desktop app for downloading and chatting with local models. Bionic is LM Studio's separate agent app, which can use local or hosted models. Our article on what is local in LM Studio and Bionic explains the difference.
The four releases
| App and version | Released | Main changes |
|---|---|---|
| LM Studio 0.4.27 | 29 October 2026 | Memory controls for the Splash engine on Mac, M5 and Radeon 8065S detection fixes, more reliable LM Link transfers |
| LM Studio 0.4.28 | 31 October 2026 | Mac update fix; chat-format fix for the Flint model |
| Bionic 1.1.10 | 29 October 2026 | oMLX engine, built-in terminal, on-device Read Aloud on Windows and Linux |
| Bionic 1.2.0 | 5 November 2026 | oMLX 0.1.0, the same Mac update fix, faster file suggestions |
If your Mac keeps reopening the old version
The changelog for LM Studio 0.4.28 says that on Mac, "Restart to apply update" reopened the current version without installing the update. Bionic 1.2.0 lists the same fix.
The fault sits in the version you are updating from, so the in-app update may fail one more time. If it does, download the current installer from LM Studio's website and install it over the existing app. Your downloaded models are stored separately from the app. If you have moved them, moving local model storage shows where to check.
What LM Studio 0.4.27 changed
On Mac:
- Splash memory controls. Splash is the fast engine for Qwen3.8 that LM Studio added in September. Version 0.4.27 adds a maximum-memory setting and a Neural Engine Prefill control for Splash models. A memory ceiling is useful on a Mac where the model shares unified memory with macOS and your other apps. Our Qwen3.8 and Splash setup article covers the engine.
- M5 Macs. The release fixes compatibility checks for engine versions that run only on M5 chips. If LM Studio offered your Mac an engine it could not run, or hid one it could, this is the fix.
On PCs with AMD graphics:
- AMD Radeon 8065S. LM Studio now detects this graphics chip for its ROCm engine, the software path for AMD graphics.
On every system:
- More reliable LM Link connections during large transfers. LM Link lets one of your devices use models loaded on another.
- A fix for a draft-model setting that stayed selected after switching to a model's built-in draft feature. A draft model is a small helper that speeds up a larger model.
- Fixes to Markdown formatting in chat.
What Bionic 1.1.10 and 1.2.0 changed
- oMLX engine for Apple silicon Macs. Added in 1.1.10 and updated to oMLX 0.1.0 in 1.2.0. See the next section.
- Read Aloud on Windows and Linux. Bionic can read replies aloud using a voice model that the changelog describes as on-device. Version 1.1.10 extends that from Mac to Windows and Linux and adds seven offline voices. On Macs with less than 32 GB of memory, the voice model no longer loads automatically, which leaves that memory for the chat model.
- Built-in terminal. The terminal inside the app is now in public builds, with configurable shell profiles and screen-reader support for its output.
- More model types in the catalog. Embedding models (used for document search) and decision models (which choose between options) now appear in the model catalog.
- Speculative decoding for MLX models. Version 1.1.10 adds this technique, in which a small helper model drafts ahead and the main model checks the draft. The changelog gives no speed figure.
- Clearer context guidance. The release fixes a misleading message about context limits when several requests use up a model's context at once.
Bionic and LM Studio can run together. Since Bionic 1.1.8 and LM Studio 0.4.26, Bionic uses LM Studio for local model work while LM Studio is open.
What is oMLX?
LM Studio's changelog gives one line: an oMLX engine for Apple silicon Macs. It does not say how oMLX differs from the MLX engine LM Studio already has, which models it suits, or how it affects speed or memory.
That leaves an Apple silicon Mac with several engines to choose from: llama.cpp for GGUF model files, MLX for MLX files, Splash for Qwen3.8, and now oMLX. GGUF vs MLX explains the first two.
Until LM Studio documents oMLX, a sensible approach is:
- Keep the engine that already works for anything you rely on.
- Try oMLX on one model you know well, with the same prompt and the same context length.
- Compare the answer, the time to the first word and the memory LM Studio reports.
- Write down the engine name and version with the result. A version numbered 0.1.0 is likely to change.
If a model stops loading after you change engines, switch back first, then work through fixing an LM Studio model that will not load.
What stays on your computer
These releases do not change where LM Studio runs a downloaded model: on your computer. Two points are worth checking:
- Read Aloud is described as on-device and its added voices as offline. The changelog does not say when the voice model is downloaded. If you need to work without a connection, try Read Aloud once with the network disconnected.
- Bionic can use hosted models. A hosted model processes your prompt on someone else's servers. Check which model is selected before sending private material. The local AI privacy guide explains what to look for.
Should you update?
| Your situation | What to do |
|---|---|
| Mac update keeps reopening the old version | Install 0.4.28, by downloading the installer if the in-app update fails again |
| M5 Mac, or a PC with a Radeon 8065S | Update to 0.4.27 or later |
| You run the Flint model | Update to 0.4.28 or Bionic 1.2.0 for the chat-format fix |
| You use Splash on a Mac with limited memory | Update to 0.4.27 or later and set a memory ceiling |
| Everything works and none of the above applies | Update when convenient |
New to the app? Start with downloading and installing LM Studio, or installing LM Studio on Linux.