清华大学

Gamma-World systematically solves architectural gaps in multi-agent world modeling via simplex agent encoding and sparse hub attention, achieving >40% average FVD reduction, zero-shot generalization from 2 to 4 agents, and 24 FPS real-time rollout.

Gamma-World systematically solves multi-agent world modeling via simplex agent encoding and sparse hub attention, enabling zero-shot generalization from 2-player training to 4-player inference and 24 FPS real-time rollout, with average FVD reduction >40%.

FaceWall Intelligence and partners have released BitCPM-CANN, the first 1.58-bit edge-side large model trained entirely on Huawei Ascend in China, achieving up to 6x memory savings while maintaining over 90% performance.

Researchers from Tsinghua University and Tencent ARC Lab have developed a pixel-aligned 3D generation method from single images, published at SIGGRAPH 2026, enabling high-fidelity geometry and texture reconstruction.

China showed strong performance in the latest AI top conference, with Tsinghua University challenging Stanford and MIT.

Tsinghua and Tencent Hunyuan team won the championship in the MLSys 2026 MoE model inference optimization competition, showcasing efficient inference optimization techniques.