Google Launches Cloud TPU v5p and AI Supercomputer: A Leap in AI Processing Power


The community used the DeepSeek V4.1 Flash model and Hermes Agent tool to perform AI optimization on the textures of Resident Evil 7, increasing the frame rate of the OnePlus 12R (Snapdragon 8 Gen 2) from approximately 20 FPS to 30 FPS, a 50% improvement. AI connected via USB to a Mac to monitor performance, identifying I/O and memory bandwidth as bottlenecks and suggesting texture resolution compression.
Google Unveils New Achievements in Global Language Intelligence: AI Supports Over 300 Languages, Covering 7.1 Billion People, 86% Proportion. Google States That Language Large Models Should Not Only Translate Text, but Also Understand Tone, Emotion, and Cultural Nuances; Meanwhile, It Points Out That Traditional Speech Recognition with Rigid Multi-step Pipelines Has Limitations.
On Sept 15, Shengshu Technology released the Vidu S2 video model series, featuring Vidu S2-Avatar for persistent interactive digital characters and Vidu S2-Editing for real-time input video editing. It also explores VR headset real-time spatial video generation/editing. Led by Zhang Jintao, PhD student under Prof. Zhu Jun.....
Anthropic upgraded the Small and Medium-sized Business Edition of Claude on September 15, 2026, adding 43 efficient workflows and deep integration with 27 mainstream business software, driving AI from a back-end support role to a core business driver. In May of this year, the intelligent connector and pre-set workflows were first introduced. This upgrade further expands the ecosystem and helps small and medium-sized businesses grow.
vivo launches an agent large model matrix with four core models for voice, on-device, and cloud, empowering terminal AI. It also pre-researches the BlueHeart 30B MoE on-device model to bring trillion-scale capabilities to phones, enhancing multimodal understanding, long-chain reasoning, and autonomous task execution, while advancing software and system ecosystems.....