Changelog 👾

Build 1

• Improve Gemma 4 tool call reliability
• Add OAuth support for MCP servers

Build 1

• Improve Gemma 4 tool call reliability
• Add support for Anthropic-compatible v1/messages output_config.effort (low, medium, high, max)
• Fixed a bug where deleting a chat folder would sometimes freeze the UI
• Fixed a bug where markdown Link popovers would appear at the top of the window

Build 1

• Add support for reasoning_effort and reasoning_tokens in OpenAI-compatible v1/chat/completions
• Adds a reasoning field to the /api/v1/models API response, indicating each model's supported reasoning capabilities/REST configuration options learn more
• Fixed a bug where Insert in chat input would sometimes not work after toggling assistant and user mode
• Fixed a bug where surrounding spaces in tool call parameters would be stripped for models that uses XML/XML-like tool call formats
• [CUDA] Fixed issue where some VRAM would not be deallocated under certain conditions
• Fixes a bug where setting reasoning to low when using Nemotron 3 Super via the /api/v1/chat or OpenAI-compatible /v1/responses API would error out

Build 4

• Fixed a bug when trying to edit an empty chat message

Build 3

• Fixed unintended parsing of tool calls within reasoning blocks
• Fixed a bug where parallel tool calls would fail for some models (for example, GLM)
• Fixed a bug in OpenAI-compatible /v1/responses which sometimes caused "Output items missing; timeline invariant violated"
• Fixed $...$ parsing so plain currency/text (for example $10, $1.23, $37 trillion) is no longer incorrectly rendered as math
• Fixed single-dollar boundary handling to reduce false positives when $ appears in normal prose
• Fixed \[ \] and \( \) handling so bracket/paren math parses correctly, and empty forms stay visible as literal text
• Fixes responsive handling of models table on narrow screens, adds resizable column handles
• Fixed a bug where Anthropic-compatible /v1/messages API would error when properties were not provided for a tool input schema
• Expose Models Directory selector in Settings
• Fixed tool calling parsing bugs for Qwen 3.5 and GLM models
  • Tool call parameters with string type were sometimes incorrectly parsed as object/number/boolean
• Added tool call grammar for gpt-oss models using llama.cpp engine, significantly increasing tool call success rate for these models (requires llama.cpp engines updated to v2.7.1 or later)

Build 2

• Global chat search now takes into account chat titles
• Add notification UI when LM Link versions are incompatible between devices
• Fixed a bug creating a duplicate onboarding popover on the LM Link page
• Make XML-like tool call parsing (e.g., Nemotron 3) more reliable for boolean values
• Fixed a bug where clicking the Attach File button in chat input would lock the text input UI
• Fixes a bug where
  tags were showing as text in markdown tables
• Fixed a responsive UI overlap bug on server page stacked content
• Fixed a bug where an unnamed chat title would appear as the chat id in the chat sidebar search results
• Fixed a bug where on certain devices, the app would crash if an image is fed to a vision model
• Fixed a bug where model load guardrails and resource usage estimates were inaccurate for some models
• Anthropic-compatible /v1/messages API now surfaces errors when the model generates an invalid tool call, enabling Claude Code to recover gracefully

Build 1

• New default: "separate reasoning_content and content in API responses" is now ON by default in order to improve compatibility with /v1/chat/completions clients
  • If your use case requires this setting to be off (previous default), you can disable it in the Developer Settings
• Fixed app header nav button hotkeys
• Add parallel parameter to /api/v1/load endpoint
• Add presence_penalty sampling parameter
• Fix hover effect visual bug on Model Picker model options in chat input
• Fixed responsive UI styling on the LM Link page
• [Linux] Fix regression caused by some app files having a space in their name.
• Fix OpenAI-compatible /v1/responses endpoint erroring on none and xhigh reasoning effort
• Fixed a bug where /v1/responses responses included logProbs for MLX models even if message.output_text.logprobs was omitted

Build 1

• ✨🎉 Introducing LM Link
  • Connect to remote instances of LM Studio, load your models, and use them as if they were local.
  • End-to-end encrypted. Launching in partnership with Tailscale.
• Fixed a bug where auto update will sometimes not work due to failing to exit before updater runs.
  • This fix only applies, when updating to the next version. That is, you might still encounter issues as you are updating to 0.4.6+1.
• Fixed Qwen3.5 RAG jinja rendering bug: "No user query found in messages"
• Updated go version to 0.25.7 for LM Link.
• [DGX Spark] Enable Direct I/O to improve model load latency
  • Requires llama.cpp engine 2.5.1 or greater.

Build 2

• Fixed a bug where LM Link connector was not included in in-app updater

Build 1

• ✨🎉 Introducing LM Link
  • Connect to remote instances of LM Studio, load your models, and use them as if they were local.
  • End-to-end encrypted. Launching in partnership with Tailscale.
• Improved tool calling support for the Qwen 3.5 model family
• Fixed a bug where loading model would sometimes fail with "Attempt to pull a snapshot of system resources failed. Error: 'Utility process is not defined'".
• Fixed a bug where autoscrolling new message behavior was not respected when clicking the Generate button
• Hides the Generate button when editing a message to avoid accidental click

Build 1

• ✨🎉 Introducing LM Link
  • Connect to remote instances of LM Studio, load your models, and use them as if they were local.
  • End-to-end encrypted. Launching in partnership with Tailscale.
• Improved tool calling support for the Qwen 3.5 model family
• Fixed a bug where loading model would sometimes fail with "Attempt to pull a snapshot of system resources failed. Error: 'Utility process is not defined'".
• Fixed a bug where autoscrolling new message behavior was not respected when clicking the Generate button
• Hides the Generate button when editing a message to avoid accidental click

Build 1

• Fixed a bug where thinking tags were not being emitted correctly through v1/chat/completions OpenAI-compatible REST API
• Fixed a bug where RAG will sometimes get stuck in "Deciding how to handle the document(s)..."
• Fixed a bug where previously detected GGUF models that were not being properly identified are now correctly recognized

Build 2

• Fixed bug causing TypeError: Cannot read properties of undefined (reading 'backendInfo')

Build 1

• Fix error when using Claude Code with LM Studio: "Invalid discriminator value. Expected 'enabled' | 'disabled'"
• Failed downloads will now automatically retry with exponential backoff. This only applies to new downloads created after this update, so existing downloads will not be affected
• Fix certain GGUFs being detected incorrectly, resulting in ?? arches and limited context lengths. Impacted unsloth/Qwen3-Coder-Next
• Fix Step 3.5 reasoning parsing
• Fix a cropped border visual bug on image attachments
• Fixed a bug where the dropdown for selecting a model file sometimes cannot be opened
• Fixed streaming REST endpoints sending headers after errors in /v1/responses and /v1/chat

Build 2

• Introducing Parallel Requests with MLX! 🎉
  • Added support for continuous batching in mlx-engine 1.0.0
  • At the moment this capability is text-only, with VLM support in the works
• Fixed a scroll bug on plugin lists
• Fixed a responsive layout bug when role and insert buttons were displayed in chat input
• Introduced new styling for message file attachments

Build 1

• Fixed a bug where Qwen3-Coder-Next could error due to an unsupported jinja template safe filter
• Fixed a bug where deleting a conversation with attachments will sometimes cause file related app operations to fail until restarted
• Fixed a bug causing Cannot read properties of null (reading 'visionAdapter')
• Fixed a bug where selected options in some interactive lms commands were not colored correctly
• Fixed a bug where large pastes in lms chat would not preserve newline formatting