Ollama, the tool for running large language models (LLMs) locally, has released version v0.34.4-rc0 today. This release candidate addresses intermittent "model not found" errors, aiming to improve the overall stability of local LLM operations. The update also incorporates version bumps for its core machine learning backends, including llama.cpp and MLX. Performance enhancements are part of this update, with specific changes to dynamically select Gemma 4 image resolution and speed up Qwen 3.8 prompt processing, particularly when utilizing MLX. Additionally, the release refines the application of structured outputs in a single pass for "thinking models," focusing on greater reliability and efficiency for various language models on local hardware.