A team from Tsinghua University and MindSpore has published new research in *Nature Machine Intelligence*, proposing the existence of a 'Compaction Law' in the evolution of large language models (LLMs). This law, interpreted through comprehensive data, provides guidance for future model development and is expected to promote optimization in LLM parameters and computational resource structures.

Meta is developing a new AI model codenamed 'Avocado,' seen as the successor to Llama, with plans to launch it in the first quarter of 2026. It is expected to be Meta's proprietary model, drawing industry attention and continuous follow-up.

Meta AI has partnered with eight major media outlets to launch a real-time news push and direct external link functionality. This aims to improve the information accuracy and influence of the Llama 4 model and initiates a reputation recovery campaign, strengthening its position and commercialization capabilities in the generative AI field.

Anthropic and Accenture have launched a multi-year partnership, planning to train approximately 30,000 professionals to accelerate the application and promotion of Claude and Claude Code among enterprise developers. They will also jointly develop AI industry solutions for highly regulated sectors, driving the enterprise adoption of large AI models.

Latest data shows that open-source AI models and solutions now account for 30% of the total global AI usage. Driven by community and enterprise efforts, they are gradually permeating numerous fields such as research, industry, and content generation, promoting AI democratization and innovation diversity.

In July 2025, major domestic and international large language models and applications saw competitive updates. Domestically, Kimi released the K2 general-purpose model, while new products like Grok 4 and GLM-4.1V-Thinking launched. Open-source Enrie 4.5 joined the fray, and Grok 4 ranked fourth in the mainstream benchmark LiveBench, reflecting the ongoing changes in the global large model competition landscape.

A new survey from the Pew Research Center indicates that 28% of US teenagers use AI chatbots daily, including ChatGPT and Google Gemini. ChatGPT leads in popularity, showing AI's deep integration into teens' daily study and communication patterns.

In November 2025, Google announced the release of the Gemini 3 model and upgraded its cloud AI infrastructure. This aims to support higher-complexity tasks and the development of multimodal AI, actively participating in the global AI infrastructure race.

Vendors represented by CrowdStrike continue to advance the concept of 'agentic Security Operations Centers (SOCs),' using AI to automate complex security processes and respond to threats in real-time, driving the upgrade of traditional enterprise security capabilities.

AI data center companies continue to acquire assets related to Bitcoin mining, leveraging the existing compute power base to accelerate AI infrastructure expansion. This is driving a concentrated migration of computing resources towards AI applications, altering the global distribution and supply-demand landscape of compute power.