On October 13, JD Industrial released an industrial world model at the 26th China International Industry Fair, targeting manufacturing, equipment maintenance, logistics and quality inspection scenarios, and pushing industrial AI from centralized analysis to on-device real-time decision-making.
Kaos IOT Technology has globally launched an industrial world model and a full-stack agent matrix aimed at scaling industrial AI adoption. The underlying Tianzhi industrial large model, validated at Kaos and Haier interconnected factories since September 2023, has accumulated over 4,700 industrial mechanism models, more than 200 expert algorithms and 110+ agent development tools, and has been deployed in over 40 scenarios across nine industries.
During the opening ceremony of the General World Model Beijing Key Laboratory, GigaAI founder and CEO Huang Guan said artificial general intelligence must go beyond language intelligence and also understand, generate, predict, and act in physical 3D space, which depends on breakthroughs in general world models. He introduced GigaAI's L1-L5 tiered system for general world models and noted that world models are relatively mature at L1 and show promise at L2 but have not yet broken through, still lagging language models overall.
Yann LeCun, founder and chairman of AMI Labs, said in Seoul that the next breakthrough in AI will come not from large language models but from systems that can understand and predict the physical world. He advocated developing "world models" to learn physical laws and causality, noted the limitations of LLMs in understanding reality, and used robots, L5 autonomous driving, and teenagers learning to drive to highlight the gap between language understanding and real-world understanding.
On October 8, Odyssey released the Odyssey-3 series of foundation world models in Standard and Pro editions. Powered by an autoregressive diffusion transformer architecture, it supports first-person and third-person navigation and independent camera movement. The Pro edition scored 66.1 on the Physics-IQ Verified video-to-video benchmark, setting a record, and ranked first in all three environment categories on WorldMark.
Fei-Fei Li called for stronger AI oversight by independent institutions and the public sector, arguing that safety assessments should not be left solely to developers and that government, industry and academia should set standards together. She also noted that World Labs is developing world models that can generate and reconstruct 3D environments for robotics training and applications in design and healthcare.
PhAI Labs, together with Oxford, Stanford, Princeton, CUHK and other teams, released JEPA-Anything, a cross-domain framework for learning predictive models across different "worlds". Its core mechanism OPF (Orthogonal Predictive Factorization) divides target states into complementary subspaces handled by independent prediction branches, and the framework was validated across vision, biology, clinical, control, molecular, physics and weather systems.