Kimi K2.5 is the latest open-source model that builds on its predecessor, Kimi K2, by leveraging pretraining on around 15 trillion mixed visual and text tokens. This multimodal model offers advanced coding and vision capabilities, including the ability to self-direct an agent swarm of up to 100 sub-agents, which can execute workflows simultaneously across 1,500 tool calls. This approach can reduce execution time by as much as 4.5 times compared to traditional single-agent setups. The model excels particularly in front-end development, capable of transforming basic conversations into fully functional front-end interfaces with interactive features. It also enhances coding tasks by reasoning over images and videos, improving the process of generating code from visual inputs. Kimi K2.5's performance has been evaluated through benchmarks like HLE, BrowseComp, and SWE-Verified, showing significant cost-effective results in various coding tasks. With Kimi Code, users can integrate this model into their existing coding environments, benefiting from its visual debugging capabilities and seamless handling of both text and graphical inputs. The introduction of K2.5 Agent Swarm marks a shift in how tasks are executed. Utilizing Parallel-Agent Reinforcement Learning (PARL), it dynamically orchestrates tasks without predefined roles, which leads to improved performance on complex projects. The agent swarm architecture not only enhances efficiency but also allows for a more flexible approach to problem-solving. In internal evaluations, K2.5 Agent Swarm has demonstrated an 80% reduction in end-to-end runtime for complex tasks, making it a powerful tool for a range of professional applications.