Moonshot AI's official Kimi Weibo account announced today that its first trillion-parameter multimodal model, Kimi K2.5, will retire at the end of this month.
In January this year, Moonshot AI announced and open-sourced its latest Kimi K2.5 model. At the same time, the Kimi assistant K2.5 version officially launched. The K2 model previously displayed in the chat interface on the official website was automatically switched to the K2.5 version.
Moonshot AI said the model was its most intelligent model at the time, achieving open-source state-of-the-art performance in Agent, coding, image, video, and a range of general intelligence tasks. It was also Kimi's most versatile model to date, featuring a native multimodal architecture that supports visual and text input, thinking and non-thinking modes, and both conversational and Agent tasks.
Kimi K2.5 is designed as a general-purpose multimodal model that can switch between conversations, complex reasoning, and automated tasks.
In July this year, Moonshot AI officially open-sourced the Kimi K3 model. It is a 2.8-trillion-parameter model built on the KDA hybrid linear attention mechanism (Kimi Delta Attention) and Attention Residuals technology. It natively supports visual understanding and has a 1-million-token context window.
Moonshot AI said Kimi K3 is the first open-source model to reach a scale of 2.8 trillion parameters. For 9 of the past 12 months, Kimi models have remained at the upper limit of the scale of open-source models.
