Ollama v0.40.15-rc0 adds MiMo V2.6 support and Linux CUDA library improvements

Ollama has released the v0.40.15-rc0 pre-release, introducing key updates to its local AI runtime capabilities. This release focuses on enhancing the integration of CUDA libraries on Linux systems and expanding the library of supported model architectures.

Users running local large language models (LLMs) on Linux will benefit from improved library handling within the MLX framework. Additionally, the update introduces support for the MiMo V2.6 model, allowing for broader experimentation with diverse model architectures through the Ollama interface.

Technical improvements in Ollama v0.40.15-rc0

• MLX/Linux Integration: The update enables the MLX backend to load bundled CUDA libraries in-process on Linux systems, aiming for more streamlined execution.
• Model Compatibility: The release adds official support for the MiMo V2.6 model architecture, expanding the range of local AI models available to users.
• Release Status: This version is currently a release candidate (rc0), intended for testing new features and stability before a full stable rollout.

Why it matters

By improving how CUDA libraries are handled in-process on Linux, Ollama enhances the stability and efficiency of local AI workloads on NVIDIA-powered hardware. The addition of new model support ensures that developers can deploy emerging architectures without needing to manually configure complex runtime environments.