From October 19 to 22, the 2026 Physical AI and World Model Conference, themed “Modeling the Physical World, Enabling a Better Future with Intelligence,” was held at the Nanjing International Expo Center. Focusing on cutting-edge fields such as Physical AI and world models, the conference brought together representatives from industry, academia, and research institutions to explore technological innovation in AI and its integration with industry.

As artificial intelligence progresses beyond understanding language and knowledge toward perceiving, modeling, and simulating the real world, enabling AI to comprehend complex environments, adapt to dynamic changes, and continually evolve through interaction with the physical world is becoming a key direction of technological exploration.

At the conference, Unisound presented U2-Flash, its full-stack intelligent healthcare insurance technology system, the AI Magic Token Global Expansion Ecosystem Hub, and its medical world model. Spanning advances in foundation model capabilities, intelligent applications in real-world industries, and the modeling and simulation of clinical processes and future care pathways, these innovations highlighted Unisound’s efforts to move AI from understanding reality to serving it and exploring future possibilities.

Model Innovation | U2-Flash: A Next-Generation High-Performance Model for Real-World Tasks

As AI becomes increasingly embedded in real-world productivity scenarios, competition among large models is expanding beyond knowledge and reasoning capabilities to encompass complex task execution and tangible application value.

At the conference, Unisound showcased U2-Flash, a next-generation high-performance model designed for real-world tasks, highlighting its latest advances in model performance, inference efficiency, and complex task handling.

Built on the U2 large model, U2-Flash follows a technical approach centered on systematic post-training and Reinforced Self-Improvement (RSI). Through a closed-loop process of “capability discovery – data generation – reinforcement training – performance validation,” it continually explores the boundaries of model capabilities, strengthens reasoning, planning, and execution in complex tasks, and advances from knowledge comprehension toward autonomous problem-solving.

Architecturally, U2-Flash adopts a sparse mixture-of-experts (MoE) design to balance performance and efficiency through higher intelligence density. By improving its ability to handle complex tasks while maintaining inference efficiency and fast response times, the model provides a foundation for real-world productivity applications with the combined advantages of greater capability, faster performance, and lower cost.

Industry Applications | Intelligent Healthcare Insurance: Turning Industry Data into Autonomous Intelligence

As the “AI Plus” initiative advances, large models are being integrated more deeply into core industry operations. A key challenge for vertical AI adoption is how to turn domain data, specialized knowledge, and business experience into intelligent capabilities that can evolve continuously while safeguarding data security and customer control.

At the exhibition, Unisound introduced its customer-sovereign large-model technology system for the healthcare insurance sector. It demonstrated an approach to domain-specific model training and continuous optimization grounded in specialized data and guided by evaluation, supporting the evolution of general-purpose models toward healthcare insurance expertise.

Drawing on its extensive experience in healthcare and medical insurance, Unisound integrates insurance knowledge, medical knowledge, and data from actual business scenarios. Targeted capability training and benchmark evaluations enhance the model’s understanding of domain knowledge and complex business rules. Multitask joint training, reinforcement learning, and expert feedback further improve its overall performance.

A data flywheel also channels questions and feedback from real-world applications into data enrichment, model optimization, and performance validation. This establishes a closed loop of continuous learning and iterative improvement, allowing industry data and business experience to accumulate as reusable domain-specific intelligent capabilities.

Industry Ecosystem | AI Magic: From Local Empowerment to Global Connectivity

Industrializing artificial intelligence requires not only continuous technological innovation but also an open, collaborative ecosystem that brings AI breakthroughs to more enterprises, entrepreneurs, and global markets.

Unisound showcased its AI Magic Token Global Expansion Ecosystem Hub, the first such hub in China dedicated to BRICS countries. The exhibit highlighted its efforts to build an AI industry ecosystem for BRICS markets and expand internationally, exploring new ways to align Chinese AI technologies with global industrial needs.

At its core is an integrated “N-G-O” ecosystem model, designed to coordinate local industry empowerment, international market expansion, and AI venture incubation.

N – Native: Empowering Local Industries with AI Agents. Through an enterprise AI gateway, an AI capability aggregation platform, and other technical services, this pillar lowers the barriers to enterprise AI adoption and supports intelligent transformation across local industries.

G – Global: BRICS Expansion Hub. Through overseas AI service locations, this pillar connects Chinese AI innovation with international market demand and helps AI technologies, products, and services enter global markets.

O – OPC: Incubating AI Creators. This pillar provides AI entrepreneurs with one-stop support spanning technology, essential resources, and business guidance, nurturing an ecosystem for AI innovation and entrepreneurship.

From empowering local industries to connecting global markets, AI Magic is exploring a new model of coordinated development across technology, industry, and ecosystems, building a bridge for Chinese AI innovation to reach the world.

Frontier Research | Medical World Models: From Understanding Life to Simulating Its Possible Futures

Unisound also brought its frontier research on medical world models to the conference. Centered on “Clinical Process Simulation Based on Synthetic Diagnostic and Treatment Trajectories,” the exhibit outlined a technical approach extending from patient-state modeling to the simulation of future care pathways, revealing new possibilities for AI to understand complex clinical processes and support continuous clinical decision-making.

Unlike conventional medical large models, which primarily focus on medical knowledge comprehension and question answering, medical world models pay greater attention to how patient states evolve over time in response to disease progression and clinical interventions. Unisound is exploring how multimodal clinical information—including medical records, laboratory tests and examinations, medical imaging, and treatment feedback—can be reconstructed into learnable diagnostic and treatment trajectories. By linking patient states, clinical actions, and subsequent outcomes, this approach provides a foundation for simulating how different clinical decisions may shape future patient conditions.

During the conference, Liu Shengping, Chief Scientist of Large Models at Unisound, delivered a keynote presentation on medical world models. He shared the company’s research and perspectives on clinical process modeling, future care pathway simulation, and the coordinated evolution of medical world models and diagnostic and treatment agents, opening up new possibilities for applying AI in complex healthcare settings.

The evolution of artificial intelligence begins with understanding the world—and its boundaries will continue to expand through sustained interaction with reality.

From breakthroughs in foundation model capabilities, to deployment and feedback in industry settings, to modeling and simulating complex clinical processes, Unisound is pursuing a path in which technological innovation and industrial applications reinforce one another. Its work explores how AI can advance beyond understanding knowledge and solving problems toward comprehending dynamic processes and simulating possible futures.

Let AI take root in reality and keep evolving as it learns to understand the world. Let intelligence reach beyond what is known and create new possibilities by exploring the future.