Kimi is a multimodal AI assistant developed by Moonshot AI, a Beijing-based AI company. Powered by the Kimi K2.5 model released in January 2026, it is built on a 1 trillion parameter Mixture-of-Experts (MoE) architecture with 32 billion active parameters per request, making it highly efficient despite its massive scale. Kimi's standout feature is Agent Swarm technology, trained with Parallel-Agent Reinforcement Learning (PARL), which enables the model to coordinate up to 100 specialized sub-agents working simultaneously across up to 1,500 coordinated steps. This parallel approach reduces execution time by up to 4.5x compared to single-agent setups. The platform operates in four modes: Instant for quick responses, Thinking for deeper reasoning, Agent for autonomous task execution, and Agent Swarm for complex multi-agent workflows. Kimi excels at visual agentic intelligence, generating code from visual specifications like UI designs and video workflows, and autonomously orchestrating tools for visual data processing. Deep Research capabilities allow Kimi to gather information from multiple sources, examine topics from different angles, and synthesize findings into visualized reports with charts and structured insights. The platform also offers specialized tools including Kimi Code for programming tasks, document processing, presentation creation, and spreadsheet capabilities. On benchmark tests, Kimi K2.5 achieved 50.2% on Humanity's Last Exam at 76% lower cost than Claude Opus 4.5, and outperformed GPT-5.2 Pro on BrowseComp for research and information retrieval. The Kimi K2.5 model is also available as an open-weight release on HuggingFace and GitHub.