The next flagship from Moonshot AI was unearthed by developers themselves. On September 28, multiple developers noticed the identifier "kimi-k3-1" appearing in the backend model registry of Moonshot AI's API platform. This identifier not only appeared in the list but also passed the interface probe and could indeed be called. Later that evening, the official Kimi open platform also posted a preview of the K3.1 model, or codenamed "K311111," stating support for a context window of up to 1 million tokens.

From the leaked configuration, Kimi K3.1 is expected to reach a maximum context window of 1 million tokens and offers three adjustable levels of reasoning intensity: Low, High, and Max. More notably, it may introduce native Agent intelligence mode and Swarm multi-agent collaboration, along with task modes such as search and batch processing. Developers have analyzed that splitting the reasoning intensity into multiple levels might correspond to different computing resource allocations and billing models—users can choose Low for cost savings or Max for optimal quality, upgrading from "a single model working alone" to "a model commanding a group of models."

Price details remain undisclosed for now. The Kimi open platform shows that the API pricing of K3.1 is the same as the current K3, although this might just be a placeholder since the model has not been officially announced. Regardless of the final pricing, from the detection of the backend identifier being adjustable to the official platform's proactive announcement, K3.1 has already crossed the threshold of recent releases—when Moonshot AI includes a million-token context, three levels of reasoning intensity, and multi-agent collaboration in its configuration, the focus of the next round of large model competition is shifting from "who is better at chatting" to "who can handle tasks on their own."