At the 2026 World Artificial Intelligence Conference and the High-Level Meeting on Global AI Governance, Zhixiang Future showcased several key industry achievements. The company unveiled the world's first multi-modal content creation agent with unlimited duration capabilities, named vivago R1. It also announced the formation of the "Belt and Road" Token Global Alliance in collaboration with institutions like CAS Brain-Like Intelligence, and jointly initiated the "Physical Intelligence Innovation Consortium" with partners including Feijie Kesi.
At a forum hosted by the China Academy of Information and Communications Technology, Mei Tao, Founder and CEO of Zhixiang Future, delivered a keynote speech titled "Towards a World Model: Native Full-Modality Drives Agent Capability Leap." He systematically outlined the development trend of large models and intelligent agents advancing in tandem towards a world model. He proposed a methodology and breakthrough path for AI to land in industries through agents, moving from the digital world to a fusion with the physical world.
In his speech, Mei Tao traced the industry's evolution: "Before 2023, agents were largely confined to academic circles. Starting in 2024, they experienced explosive growth, bridging the gap between large models and practical applications. In the future, agents will achieve self-evolution, iterating from single agents to multi-agent clusters, growing into digital employees capable of handling complex tasks and delivering standardized results."
He further proposed that large models are responsible for enabling AI to understand and perceive the world, while agents empower AI with the ability to execute and act. A native full-modality world model integrates cognition, action, and feedback into a closed loop, propelling AI from content generation to altering the real world. Mei Tao also emphasized the industry's developmental baseline, stressing the need to continuously enhance agents' inherent safety capabilities to mitigate various application-derived risks and ensure the safe and controllable deployment of AI.
Focusing on the Direction of AI Industry Evolution: Symbiotic Synergy Between Foundation Models and Agents
Mei Tao believes that current large models are leaping forward at an unprecedented pace, with top-tier models reaching human genius-level intelligence. However, a significant industry gap remains in translating foundational models into effective real-world applications, characterized by deployment difficulties, weak adaptation, and lack of control. Agents possess core capabilities such as autonomous memory, logical planning, tool orchestration, and multi-end collaboration. They can transform the generative and reasoning advantages of foundation models into standardized, verifiable, and deliverable systematic industry capabilities. The combination of foundation models and agent capabilities is becoming the core paradigm for AI to empower all industries.
He further summarized four key technological evolution paths for agents: First, capability abstraction upgrades, shifting from basic tool invocation to accumulating reusable industry skills, treating industrial experience as fundamental agent invocation units. Second, engineering paradigm iteration, moving from singular prompt optimization to ensuring system stability and low-cost, continuous output. Third, iteration mechanism innovation, upgrading from manually preset fixed workflows to feedback-driven autonomous evolution workflows. Fourth, collaboration framework upgrades, progressing from single-agent independent operation to multi-agent collaborative orchestration and scheduling.
Mei Tao further pointed out that in complex tasks and scenarios, single agents have clear capability boundaries and cannot independently complete cross-chain, high-precision creative tasks. Only through the collaborative work of multi-agent clusters, simulating the division of labor in professional teams, can they "handle" complex industrial challenges. To ensure the stable, efficient, and orderly operation of multi-agent clusters, an AgentOS (Agent Operating System) is an indispensable core infrastructure.
Mei Tao defines AgentOS as the "last mile" in unlocking the value of large models. It is not merely a collection of tools but a reusable, intelligently orchestrated system of creative productivity components, serving as the underlying foundation that supports stable output and long-chain reasoning for multi-agents.
To this end, Zhixiang has independently developed a new HD-AgentOS. Relying on a three-layer coupled architecture of resource, system, and capability layers, it constructs a complete support system for agent industrial deployment. The resource layer provides foundational resources and atomic capabilities that agents can invoke, building the execution foundation. The system layer is responsible for the full-process operation, maintenance, risk control, monitoring, and security governance of agents. The capability layer encapsulates models, tools, knowledge, and processes into domain-specific skills.
Major Launch of vivago R1: World's First Unlimited-Duration Creation Agent
Leveraging core technological accumulations in multi-agent collaboration and AgentOS, Zhixiang Future officially launched the world's first unlimited-duration creation agent, vivago R1, at WAIC 2026.
This is the world's first multi-modal creation agent with unlimited-duration video generation and editing capabilities. It has evolved from assisting in generating single-point materials to becoming an "intelligent partner" capable of autonomously planning, orchestrating, and completing long-chain creative tasks. The core advantages of vivago R1 can be summarized by three key traits: "Long Duration, Long-Term Thinking, and High Stability," directly addressing three long-standing industry pain points of duration limitations, logical inconsistencies, and unstable output, which align with Mei Tao's proposed core requirements for multi-modal creation.
First, unlimited duration with full-scenario adaptation. Current mainstream AI video products in the industry typically only support 15-30 second short clip generation, failing to meet long-cycle commercial creation needs. vivago R1 completely breaks the industry's duration barrier, supporting continuous and coherent generation of videos of any length, capable of producing minute-level finished videos within half an hour. It fully adapts to long-cycle creation scenarios across all categories, including short dramas, documentaries, brand promotional videos, film/TV finished products, and overseas social media drama accounts, filling the market gap for long-video creation. Simultaneously, the product possesses dual unlimited capabilities in content breadth and parallel generation, supporting multi-threaded batch production of differentiated materials to suit the scaled production needs of content teams.
Second, long-task thinking ensures coherent and unified creative logic. Long-video creation is not simply splicing short clips together; it requires complete narrative logic, consistent visual style, stable character IPs, and scene settings. vivago R1 possesses long-task thinking ability, enabling it to autonomously complete refined logical conception, shot arrangement, and style calibration. This effectively solves industry challenges like jarring scene transitions, character breakdowns, narrative disconnects, and stylistic inconsistencies in AI-generated videos, ensuring the integrity and professionalism of the entire production.
Third, high-stability generation empowers commercial-scale delivery. Traditional large models suffer from probabilistic output deviations and low usability rates for finished videos, significantly increasing corporate debugging costs. Relying on the full-process orchestration and governance capabilities of AgentOS, vivago R1 achieves controllable and calibratable creation across the entire chain. It increases the success rate of effectively usable content to 85%, far above the industry average. This drastically reduces the time and labor costs associated with repeated modifications and debugging. The generated content can be directly applied to commercial production, brand communication, and market delivery, moving beyond the industry pain point of repeated "drawing lots" and rework.
In persistent, long-chain creation scenarios like social media content and commercial marketing, a seemingly simple "storyboard generation" task requires not only generating visuals based on a script but also understanding cinematographic language, narrative pacing, emotional expression, and even distinguishing the different requirements for storyboards across genres like short-video vlogs, short dramas, and product material clips. This depth of industry understanding cannot be achieved by simply calling a large model API.
Through its multi-agent collaboration mechanism, vivago R1 atomizes and professionally orchestrates the complete creative chain: "understanding the task → conceiving the storyline → creating the storyboard script → generating core materials → generating ultra-long videos." This ensures the organic unity of narrative, shots, characters, style, sound, and aesthetics, truly achieving the leap from a "single-point tool" to a "full-chain creation system."
Strategic Synergy: Dual Alliance Initiatives Open a New Chapter in Global Ecosystem Building
Beyond the groundbreaking product launch, Zhixiang Future also advanced two major ecosystem strategic initiatives during WAIC 2026.
Zhixiang, jointly with innovative enterprises like Feijie Kesi and top scientific research institutions, co-initiated the "Physical Intelligence Innovation Consortium" joint proposal. This aims to pool strengths from industry, academia, and research to build a new pattern of industrial synergy for native full-modality world models.
Furthermore, Zhixiang formally reached a strategic cooperation agreement with institutions including CAS Brain-Like Intelligence, signing to establish the "Belt and Road" Token Global Alliance. Guided by the core concept of "Token Without Borders, AI Without Islands," the alliance aims to propel China's native full-modality AI capabilities across distances, connecting with global intelligent infrastructure, and assisting in the intelligent upgrade and digital ecosystem interconnection of countries and regions participating in the Belt and Road Initiative.
Currently, the competitive dimensions of the artificial intelligence industry have been comprehensively upgraded. The "1+1+3" business model crafted by Zhixiang represents a deep response to the new propositions of the AI industry—maintaining globally leading competitiveness in foundational models while delving deepest into the vertical tracks where users most need intelligent capabilities to be unleashed.
Mei Tao proposed that the long-term capability ceiling of an agent is determined by the underlying foundation model. The foundation model is responsible for perceiving, inferring, and reconstructing the real world, while the agent, as the execution unit, completes autonomous decision-making, tool invocation, and feedback closure. Only through the synergistic combination of these two, underpinned by a world model foundation, can the continuous iterative evolution of agents be driven.