Recently, Kimi K2.5, Visual Agentic Intelligence was released, and it is easily one of the strongest open-source multimodal models available today. If you look at the benchmarks across agents, coding, ...