Moonshot: kimi-k3

Kimi K3 is Kimi's most powerful flagship model to date, boasting 2.8 trillion parameters. Built upon the KDA hybrid linear attention mechanism (Kimi Delta Attention) and attention residuals technology, it natively supports visual understanding and features a 1 million token context window. It is the world's first open-source model with over 3 trillion parameters, designed for cutting-edge intelligent scenarios such as long-range programming, knowledge work, and reasoning.