LM Studio 0.3.16

LM Studio 0.3.16 is available now as a stable release. This version includes the public preview of community presets, automatic deletion of least recently used Runtime Extension Packs, and a way to use LLMs as text embedding models. It also introduces the lms chat command to the CLI, allowing you to chat with a model in the terminal.

Upgrade via in-app update, or from https://lmstudio.ai/download.

---

📣 An earlier build of 0.3.16 (build 6) was missing engines within the app bundle on Windows. If this was the first version you ever installed, please head to Ctrl + Shift + R to install the engines manually, or download and reinstall 0.3.16 build 7 from https://lmstudio.ai/download.

0.3.16 - Full Release Notes

• Ability to publish your presets to the LM Studio community hub and share them with others

  • Docs: https://lmstudio.ai/docs/app/presets/publish
  • Downloading public presets does NOT require a login. Publishing, liking, forking does
  • Share links with others directly, or on Discord / Reddit / X (No way to browse presets yet)

• Staff Picks polish: introduce new aggregate format that combines formats (e.g. GGUF and MLX) into a single entry

• Custom Settings for models

  • A new way for model authors to define custom capabilities for models. For example, Enable Thinking on / off for Qwen3 models
  • At the moment, only available for Qwen3 models under Staff Picks.
    • Later we'll share public docs for doing this with your custom models

• Add "Offload KV Cache to GPU Memory" option to model load options and GPU settings

• Add lms chat command to the CLI to chat with a model in the terminal (Thanks @mayfer)

• LM Studio REST API (/api/v0): return model capabilities in GET /models response

  • e.g. "capabilities": ["tool_use"]

• Auto-deletion of least recently used Runtime Extension Packs

  • Can disable in App Settings (⌘/Ctrl + ,)

• Show System Prompt button in chat top bar when sidebar is collapsed

• Use Cmd / Ctrl + Shift + D to create a duplicate of the current chat

• Use Cmd / Ctrl + E to open the System Prompt editor

• Use Cmd / Ctrl + W to close the System Prompt editor tab when it is open

• App Settings (⌘/Ctrl + ,): sections now have their own tabs for easier navigation

• Add button to access downloads panel in "User" UI mode

• Added a dropdown in the model editor (⚙️ in My Models) to allow overriding the domain type of a model

  • This allows you to use an LLM as a text embedding model, or vice versa

• Add "Reveal in Finder" context menu option on the chat sidebar body

  • Open your conversations folder in your OS native file explorer and reorg files / folders
    • Changes will be reflected live in LM Studio

• [MLX] Register chat_template.jinja as a source for chat templates

Bug Fixes:

• Fix Qwen3 jinja template parsing error "OpenSquareBracket !== CloseStatement"
• Fix nomic-bert-moe, nomic-embed-code not showing under embedding models
• Fix flickering horizontal line at bottom of generating message
• Fix inability to be create a new folder when no chat is selected
• Fix prompt processing % UI not showing
• Bugfix: [OpenAI-like REST API] Don't add empty function name string after the first packet of streamed function calls
• Bugfix: Correctly filter out unsupported models when listing downloaded models
• Bugfix: Don't assume Vulkan iGPU memory is Shared Memory
• Fix architecture badge text alignment
• Fix SmolVLM jinja render error "Expected iterable type in for loop: got StringValue"
• Fixed GTE Qwen embedding models are not recognized as text embedding
• Fixed a bug where if no engine extension pack for a specific format is installed, the indexing will fail
• Fixed a bug where some strings were parsed as [object Object] when using RAG
• Fixed a bug showing duplicate models in Load Model in tray menu
• [MLX] Fix installation issue that caused ModuleNotFoundError: No module named 'mlx_engine'
  • You can press the "Fix" button in the runtimes page if you are seeing this error with an existing MLX runtime