TensorSharp, a pure .NET inference stack, has rapidly expanded its model support, moving from DeepSeek V4 (284B parameters) to GLM-5.2 (744B) in just three weeks. This represents a 2.6x increase in capacity while maintaining token-level alignment with llama.cpp. For .NET developers, this means more options for running large language models without leaving the ecosystem. The project's focus on compatibility with established tools like llama.cpp suggests it aims to be a viable alternative for cross-platform AI deployment. While still niche compared to Python-based stacks, this progress indicates growing interest in .NET for AI workloads, particularly in enterprise environments where .NET is already prevalent. The rapid iteration also highlights the fast-moving nature of the AI model landscape, with new architectures emerging frequently.
TensorSharp expands .NET inference to support 744B parameter models, maintaining llama.cpp alignment. A sign of growing .NET presence in AI infrastructure.