On January 30, 2026, Moore Threads unveiled the open-sourcing of the TileLang-MUSA project, which has successfully concluded its functional verification and feature development phases. This initiative is designed to support the TileLang programming language and has undergone rigorous validation on locally produced GPUs, including the MTT S5000 and S4000 models. Its core objective is to minimize development hurdles by leveraging high-level abstraction and compiler optimization techniques, thereby furnishing domestic computing platforms with potent tools for AI and high-performance computing endeavors.

TileLang boasts a Python-inspired declarative syntax, facilitating automatic loop optimization and memory scheduling. Empirical tests reveal that it slashes code volume by roughly 90%, while achieving matrix operation performance that is 95% on par with manually optimized counterparts. The project's codebase is now accessible on GitHub, inviting collaboration and further innovation.