Changelog • LM Studio 0.3.19

Build 4

Fixed a bug where certain models fail to load (Cannot read properties of null (reading '1'))
Fixed "Reload last model" button showing even if no model was used previously
Fixed a bug where newly joined Hub organizations would not show up in the Publish Preset dropdown until the app is restarted

Build 3

Fixed a bug where some Mistral model's correct tool calls would fail to parse
[Linux][ROCm] Support for AMD 9000 series GPUs and Ryzen AI PRO 300 series integrated GPUs

Build 2

Button to navigate to Previous Chat, Next Chat
Auto deletion of no-longer-needed engine dependencies to save disk space
Fixed a bug where sometimes the model status in the server page will not update correctly
Fixed staff picks sometimes failing to load
Fixed a bug where chat listings panel losing scroll position when selecting a chat

Build 1

Added "Think" toggle button into chat input for models that support turning thinking on/off (e.g. Qwen3, Magistral)
Fixed a bug where runtime updates sometimes did not upgrade the model format default selection
Fixed a bug where symlinks/junctions in the model directories are not followed correctly
Fixed a bug where the usage stats in OpenAI streaming responses were incorrect
Fixed a bug where multi-GPU model loading sometimes failed with custom GGUF quantizations (e.g. Q6_K_XL)
Fixed a bug where parallel tool calls via OpenAI streaming API were not handled correctly
[Windows] Fixed a bug where sometimes start item would get reenabled after being disabled in Task Manager