LM Studio Bionic 1.1.10: Power-User Guide to the Terminal, Speculative Decoding and October Updates

Last updated: October 30, 2026

LM Studio Bionic 1.1.10, released October 29, 2026, makes the in-app terminal public, adds the oMLX engine for Apple Silicon Macs, brings speculative decoding to MLX and SafeTensors models, and adds embedding and decision models to the catalog. It follows October updates that let Bionic and LM Studio run together.

This guide is based on the official Bionic 1.1.10 changelog, the 1.1.9 and 1.1.8 release notes, and LM Studio’s Bionic documentation. It focuses on what changed in October and how a power user should configure it.

Contents

• What’s new in LM Studio Bionic 1.1.10
• October 2026 releases at a glance
• The in-app terminal
• oMLX, speculative decoding and MoE caching
• Decision and embedding models
• Context, MCP and sessions
• How to set up the new features
• Limitations
• Frequently asked questions

What’s new in LM Studio Bionic 1.1.10

LM Studio describes Bionic as the next generation of LM Studio designed for agentic work with open models. It is a separate app from LM Studio, aimed at coding, research and complex work with documents and files. Within a project you create separate sessions for separate tasks and can run several sessions in parallel.

The headline changes in 1.1.10 are:

• In-app terminal for everyone: now in public builds, with configurable shell profiles, restored sessions that start automatically when selected, and screen reader support for terminal output.
• oMLX engine: a new engine option for Apple Silicon Macs.
• Faster generation options: speculative decoding for MLX models and SafeTensors model support for speculative decoding.
• New model types in the catalog: embedding and decision models.
• Read Aloud beyond Mac: on-device Read Aloud on Linux and Windows, plus seven additional offline voices.
• Longer sessions: more reliable context compaction when conversations exceed model limits.

October 2026 releases at a glance

Release	Key changes
Bionic 1.1.8	Coexistence with LM Studio 0.4.26+, /v1/decisions and /v1/systemone endpoints, decision model controls in the app and REST API, Word, spreadsheet, presentation and HTML previews, draggable Canvas panel
LM Studio 0.4.26 (Oct 13)	Coexistence with Bionic 1.1.8+, CUDA 13 on Windows ARM, llama.cpp runtime 2.50.0
Bionic 1.1.9 (Oct 21)	On-demand MCP tool discovery, on-device Read Aloud, Neural Engine Prefill and maximum memory controls for Splash, CSV and TSV previews, session references by dragging tabs
LM Studio 0.4.27 (Oct 29)	LM Link reliability, Radeon 8065S ROCm detection, Splash memory and Neural Engine Prefill controls on Mac
Bionic 1.1.10 (Oct 29)	Public in-app terminal, oMLX engine, MLX and SafeTensors speculative decoding, embedding and decision models in the catalog, MoE expert GPU cache (experimental)

The in-app terminal

The terminal is the most visible change for people who use Bionic as a coding agent. You can now run shell commands next to the agent’s work without switching to a separate terminal app. Version 1.1.10 adds configurable shell profiles, starts restored terminal sessions automatically when you select them, and fixes keyboard shortcuts, clipboard handling, saved shell selections and IME input in the terminal.

A practical pattern: let the agent edit files, then run tests in the same window and paste failures back into the chat. If you also use a hosted coding agent, compare this workflow with our Claude Code power-user guide.

oMLX, speculative decoding and MoE caching

Three changes target local inference speed. First, the oMLX engine is a new option for Apple Silicon Macs. Second, speculative decoding, where a small draft model proposes tokens that the main model checks, now works with MLX models and with SafeTensors models. Bionic 1.1.10 also fixes stale draft-model selections when you switch to bundled MTP drafting.

Third, an experimental MoE expert GPU cache setting is now available for llama.cpp mixture-of-experts models. LM Studio labels it experimental, so benchmark it on your own hardware before relying on it.

For a comparison with another local runner, see our Ollama new features guide.

Decision and embedding models

Bionic 1.1.8 added support for an OpenAI-compatible /v1/decisions endpoint and a /v1/systemone endpoint compatible with Jev (TypeSafe AI), along with decision model load settings, activity status, cancellation controls, and inference and management through the REST API. Version 1.1.10 adds embedding and decision models to the model catalog, so you can find and download them in the app.

Decision models return structured judgments rather than free text. Our LFM3 guide covers Liquid AI’s d1 decision models, and the Perplexity Decisions API guide covers a hosted alternative.

Context, MCP and sessions

Agent sessions use context quickly, and several October fixes address that. Bionic 1.1.9 lowered context usage by discovering MCP tools on demand, made skill context more reliable across retries, copies and forks, and gave agents access to attachments across sessions. It also lets you reference another session by dragging its tab into the composer. Version 1.1.10 makes context compaction more reliable and fixes misleading context-limit guidance when several requests exhaust a model’s context.

Where a session runs is your choice. The models documentation lists three options: cloud models in LM Studio Secure Cloud with Zero Data Retention, local models on your computer, and remote models running on another device through LM Link. Bionic’s product page names GLM 5.2, Kimi K3 and DeepSeek V4 Pro as frontier open models you can use; see our Kimi K3.1 Pro guide and DeepSeek V4.1 Flash guide for related models.

How to set up the new features

1. Update Bionic to 1.1.10 or later, and LM Studio to 0.4.26 or later if you want both apps running together. Bionic then uses LM Studio for local inference while LM Studio is open.
2. Open a project, start a session for one task, and pick its model from the model picker: Cloud, Local or Remote.
3. To change the default for new sessions, open Settings, then General, and choose a Root model.
4. If the model supports configurable reasoning, choose a reasoning level for the session.
5. On an Apple Silicon Mac, open Runtime settings to review available engines and backend downloads, including oMLX.
6. To try speculative decoding, choose a draft model for an MLX or SafeTensors model and compare speed with it on and off.
7. Open the in-app terminal, set a shell profile, and use it to run builds and tests alongside the agent.
8. For decision models, download one from the catalog and call it through the /v1/decisions endpoint or the REST API.

Limitations

• The MoE expert GPU cache setting is marked experimental.
• Cloud models need an internet connection, a signed-in LM Studio account with billing set up, and available credits.
• Remote models appear only while the linked device is available and sharing its local models.
• Voice model auto-loading now defaults to off on Macs with less than 32 GB of memory.
• A local model must be downloaded and must fit your hardware, so check memory before choosing a large model.

Frequently asked questions

What is LM Studio Bionic?

LM Studio Bionic is a separate agent app from LM Studio, built for agentic work with open models such as coding, research and work with documents and files. A session can use local models, remote models over LM Link, or larger open models in LM Studio Secure Cloud with Zero Data Retention.

What is new in LM Studio Bionic 1.1.10?

Bionic 1.1.10, released October 29, 2026, makes the in-app terminal available in public builds, adds the oMLX engine for Apple Silicon Macs, adds speculative decoding for MLX and SafeTensors models, lists embedding and decision models in the catalog, and extends on-device Read Aloud to Linux and Windows.

What is the difference between LM Studio and LM Studio Bionic?

LM Studio is the app for downloading and running local models, with developer and server features. Bionic is a newer, separate agent app. LM Studio says you can keep using LM Studio alongside Bionic for advanced low-level configuration, and since LM Studio 0.4.26 and Bionic 1.1.8 the two apps can run at the same time.

Does LM Studio Bionic cost money?

Local and remote models run on your own hardware. Cloud models in LM Studio Secure Cloud require a signed-in LM Studio account with billing set up and available personal or organization credits, and their usage consumes those credits.

Does LM Studio Bionic support MCP and skills?

Yes. Bionic 1.1.8 fixed MCP connection status in chats and gave skills their own composer suggestions category, and Bionic 1.1.9 lowered context usage by discovering MCP tools on demand and made skill context more reliable across retries, copies and forks.

Sources

• LM Studio: Bionic 1.1.10 changelog (October 29, 2026)
• LM Studio: Bionic 1.1.9 changelog (October 21, 2026)
• LM Studio: Bionic 1.1.8 changelog
• LM Studio Docs: Choose a Cloud, Local, or Remote Model
• LM Studio: LM Studio app changelog

Image: official LM Studio logo from the LM Studio brand assets page, used under its usage terms, placed unchanged on a white background. LM Studio is a trademark of Element Labs. FreeAnswers is not affiliated with or endorsed by LM Studio.

Related guide: hf CLI new features guide: huggingface_hub 2.3 and 2.4, Jobs, Sandboxes and Buckets.

Related answers

• Codex CLI Update 0.166: Power-User Guide to October 2026 Features
• Ollama New Features: Power-User Guide to Speech, Decision Models and October 2026 Releases
• Gemini 4 Argon: Power-User Guide to Access, API, Antigravity and Limits