The latest release candidate of Ollama, v0.32.3-rc0, arrives with a focused set of refinements that strengthen the platform’s stability and polish for self-hosted AI users. Rather than introducing flashy new features, this update fine-tunes existing components to ensure a smoother, more reliable experience.
Owners of Apple Silicon machines will appreciate the MLX backend update, which brings Ollama in line with the latest optimizations for on-device performance. This means models running via MLX should see better efficiency and responsiveness, without any user intervention beyond the update itself.
For those working with GLM-based models, a parser fix now properly finalizes incomplete tool calls, preventing silent failures or truncated outputs that could disrupt automated workflows. This change ensures that tool interactions behave predictably, which is key for building reliable AI-powered applications.
Behind the scenes, the Laguna model has been aligned with the upstream llama.cpp library, maintaining consistent behavior and preventing drift as the underlying engine evolves. Finally, the project’s documentation has been refreshed to clearly indicate any deprecated features, giving administrators a heads-up when planning their upgrades.
