360 Security Technology (SH:601360) Zheng Quan Shi Bao Wang·2025-08-03 06:13Core Insights - The evolution of AI agents is accelerating, with 360's Nano AI upgrading to a "multi-agent swarm" system, marking a significant advancement to L4 level AI agents capable of collaborative tasks [1] Development Stages of AI Agents - AI agents are categorized into three development stages: - L1: Chat assistants, primarily tools for suggestions and emotional support, considered "toy-level" AI, such as GPTs [2] - L2: Low-code workflow agents, which have evolved from "toys" to "tools," delivering semi-finished products rather than final outputs [2] - L3: Autonomous planning agents, capable of reasoning, exemplified by the initial version of Manus and over 50,000 vertical expert agents in Nano AI, which can perform deep reasoning in specific tasks but struggle with cross-domain complex issues due to lack of collaborative planning [2] Multi-Agent Swarm Capabilities - The new swarm collaboration framework in Nano AI allows multiple reasoning agents to flexibly group, nest, and collaborate on complex tasks, functioning like a bee swarm [3] - The framework supports both single swarm operations and larger swarm formations, adapting to various user tasks and configurations [3] Performance and Applications - The Nano AI swarm has successfully executed over 1,000 steps of tasks continuously for 2 hours, consuming over 20 million tokens [4] - The system has created China's first "one-sentence blockbuster" multi-agent swarm, significantly reducing the time to produce complex tasks like a 10-minute film from at least 2 hours to just 20 minutes [4] - It can also generate viral short videos for e-commerce, translating content into multiple languages and creating promotional videos from simple inputs [4]