CONNECT THE DOTS

Find your next insight.

Search stories, companies and the ideas behind China’s AI.

312 storiesPage 22 of 52
Robotics

Hesai Launches Kosmo 3D World Data Engine to Solve Real-World Data Scarcity for Physical AI

Shanghai-based lidar company Hesai unveiled Kosmo, a 3D world data engine that captures physical environments and converts them into high-fidelity digital assets for robot training and world model development. Demonstrated at Kosmo AI Day on Oct. 14, the system generated over 10,000 training trajectories from a single scan and enabled a partner robot to achieve a 99% task-success rate after replacing hundreds of hours of manual modeling.

Companies

Ant Group's Health AI App and Alipay AI Payment Named to TIME's 2026 Best Inventions

Ant Group's health AI app Ant Afu (AQ) and Alipay AI Pay were both selected for TIME magazine's 2026 Best Inventions list, announced October 15. Ant Afu, with over 150 million users and connections to 5,000 hospitals, was described by TIME as expanding healthcare access in underserved Chinese regions. Alipay AI Pay, which enables AI agents to complete transactions on users' behalf, has surpassed 100 million users since its 2025 launch.

Models

Xiaomi Open-Sources MiDashengLM-Spatial, Claiming First End-to-End Spatial Audio Understanding Model

Xiaomi has released MiDashengLM-Spatial under the Apache 2.0 license, describing it as the first open-source end-to-end spatial audio understanding large model. Built on its existing MiDashengLM-7B audio-language model, the new version adds binaural perception and spatial reasoning through a dedicated spatial audio encoder, targeting smart home and car cockpit applications within Xiaomi's human-car-home ecosystem.

Chips

Moore Threads Completes Day-0 Adaptation of MTT S5000 for StepFun Step 5 Preview Model

Chinese GPU maker Moore Threads announced it has completed Day-0 adaptation of its MTT S5000 AI computing card and MUSA software stack for StepFun's newly open-sourced Step 5 Preview model. The adaptation covers mixed attention, mixed precision inference and speculative decoding, demonstrating growing domestic hardware readiness for frontier Chinese large language models.

Models

Kuaishou's Kling 4.0 Goes Live to All Users With 30-Second Generation and Workflow Integration

Kuaishou has fully launched Kling 4.0 on klingai.com, opening its latest video generation model to all users. The upgrade introduces 30-second single-shot generation, 4K 10-bit HDR output, dual-channel stereo audio, improved lip-sync accuracy, and MCP/CLI integration, signaling a strategic shift toward embedding the model into professional production workflows rather than keeping it as a standalone demo tool.