Kimi K2.5 (EU)

Kimi K2.5 is an open-source, native multimodal agentic model released by Moonshot AI, built through continual pretraining on approximately 15 trillion mixed visual and text tokens atop the Kimi-K2-Base foundation. This training approach gives the model integrated vision and language understanding, allowing it to process both visual and textual inputs within a single architecture rather than relying on bolted-on vision encoders. The model is designed to operate in both instant and thinking modes, supporting conversational and agentic paradigms depending on the complexity of the task at hand.

Beyond its multimodal foundation, Kimi K2.5 introduces a self-directed swarm architecture capable of coordinating up to 100 agents, positioning it for complex, multi-step agentic workflows rather than single-turn chat interactions. This swarm capability, combined with tool calling and structured output support, makes the model well suited for developer-facing applications that require autonomous task decomposition, planning, and execution. The open-weight release means organizations can self-host and adapt the model, while the EU regional deployment through cloud infrastructure provides a data-residency option for European use cases.