Ollama's latest release, v0.32.14, brings several practical improvements for developers running local language models. The new automatic WebP transcoding simplifies handling image inputs in multimodal models, removing a common preprocessing step. Additionally, enhanced compatibility with Qwen system messages ensures more reliable instruction following for Qwen-based deployments. The update also includes refreshed llama.cpp and MLX backends, which may yield performance gains on CPU and Apple Silicon systems. For teams integrating Ollama into their pipelines, these changes reduce friction and improve consistency across different model architectures. While not a major version bump, this release reflects the project's steady evolution toward a more robust local inference platform.
Ollama v0.32.14 adds WebP auto-transcoding and Qwen system message compatibility, plus llama.cpp and MLX updates, improving local LLM workflows.