On the evening of July 28, Moonshot AI made headlines by open-sourcing its Kimi K3 model, boasting an impressive 2.8 trillion parameters, and simultaneously releasing both the model weights and a detailed technical report. On this very same day, Moore Threads made a significant announcement: its AI training and inference integrated accelerator, the MTT S5000—powered by the MUSA software stack—had swiftly achieved compatibility and was operating stably with the Kimi K3 model.

As the world's pioneering open-source large model with 3 trillion parameters, Kimi K3 stands out with its remarkable context window of 1 million tokens, native support for visual understanding, and the incorporation of multiple architectural innovations. These features pave the way for its application in cutting-edge intelligent scenarios, such as long-range programming.

This compatibility milestone not only underscores the rapid response and full-stack collaborative prowess of Moore Threads' MUSA software ecosystem but also serves as a testament to the capability of domestic full-function GPUs in bolstering the research, development, and deployment of large models with trillion-level parameters.