GLM-5.2, released by Z.ai, represents a significant leap for open-weight models, matching or exceeding closed-source models in agent and coding benchmarks. Its release amid the ban on Claude Fable highlights economic and geopolitical implications, sparking debates on open vs. closed models.

PP-OCRv6 is PaddleOCR's latest universal OCR model family, scaling from 1.5M to 34.5M parameters across three tiers, supporting 50 languages. It delivers a +4.6 percentage point improvement in text detection Hmean and +5.1 in recognition accuracy over PP-OCRv5_server. New architecture includes PPLCNetV4 backbone, RepLKFPN for detection, and EncoderWithLightSVTR for recognition. Supports multiple inference backends: Paddle Inference, Transformers, and ONNX Runtime.

JUPITER, Europe’s first exascale supercomputer at Germany’s Forschungszentrum Jülich, runs on NVIDIA Grace Hopper Superchips and NVIDIA Quantum-X800 InfiniBand networking — and it’s had a busy year. Four projects running on JUPITER demonstrate what exascale computing can actually do: map the human brain at cellular scale, simulate the entire Earth’s climate at 1-kilometer resolution, build AI systems for next-generation wireless networks, and simulate a universal 50-qubit quantum computer.

The U.S. National Science Foundation's NAIRR pilot program has driven over 700 projects in two years, with NVIDIA providing DGX nodes and technical support. Highlights include Polymathic AI's fluid simulations, University of Michigan's fusion model for energy storage, and Boston University's BEACON pipeline for infectious disease detection.

At ISC Hamburg, NVIDIA introduced DAQIRI, ALCHEMI NIM, and cuPhoton software to accelerate science. cuPhoton achieved up to 14,900x speedup in astronomical data processing, while DAQIRI enables real-time AI on high-speed data streams. ALCHEMI accelerates materials simulation, with Lila Sciences using it to speed up materials screening by 50x.

Los Alamos National Laboratory (LANL) is building three new supercomputers with HPE and NVIDIA, powered by NVIDIA Vera CPUs, to accelerate scientific discovery and enable agentic AI for science. The systems—Mission, Vision, and Veritas—use the HPE Cray GX5000 architecture with the NVIDIA Vera Rubin platform. Early tests show Vera CPUs delivering 7x higher performance on URSA workloads and over 3x on heat transfer simulations compared to Crossroads x86 CPUs. Mission, expected in 2027, will replace Crossroads for classified national security work, while Vision will support fundamental science.

AI's growth is energy-hungry. Eco Wave Power uses NVIDIA AI and digital twins to convert ocean wave energy into clean electricity, leveraging existing coastal infrastructure. Wave energy is abundant, less intermittent, and can potentially power AI factories and data centers.