A number of developers have observed the emergence of a callable identifier, namely 'kimi-k3-1', within the back-end model registry of the Yuezhi'anmian API platform. Following this discovery, the official Kimi open platform offered a sneak peek at the K3.1 model, which boasts a context window capable of supporting 1M tokens. This upcoming model is projected to accommodate a maximum context window of up to 1 million tokens. It will provide users with three distinct levels of inference intensity to choose from and might also introduce an agent mode, multi-agent collaboration capabilities, as well as task modes tailored for search and batch processing operations. The availability of multiple inference intensity levels is likely to be linked to varying computing power configurations and corresponding billing schemes. Presently, the K3.1 API price listed on the platform mirrors that of the current K3 model; however, this figure is likely to serve as a temporary placeholder. The Kimi K3.1 model is slated for an official rollout next month.