In the past 24 hours (November 18-19, 2025), the overseas AI track has ushered in the release of Google Gemini 3 and a number of billion-dollar investment and financing, while in China, it has focused on government affairs platforms, basic scientific research and AI literacy in primary and secondary schools, showing a two-line evolution pattern of "technology upgrading + governance acceleration" as a whole.
- Google released the multimodal large model Gemini 3, directly embedded in search and core applications
Gemini 3 is positioned as Google's strongest multimodal model to date, significantly improving the benchmark of complex reasoning and code generation, and accessing Search, Gemini App and Workspace in the form of preview, marking the further "AI priority" of the search experience.
- Nanjing released the country's first white paper on generative AI literacy for primary school students
The "White Paper on Generative AI Literacy for Primary School Students in Nanjing" was officially released, giving specific suggestions from the dimensions of risk cognition, tool selection, homework and classroom application, and is regarded as the first generative AI normative document specifically for primary school students in China.
- Microsoft and Nvidia plan to invest up to $15 billion in Anthropic, increasing generative AI infrastructure
The two companies announced plans to invest up to $15 billion in Anthropic over the next few years, which will be mainly used to train larger-scale models and expand computing power, which is regarded as another typical case of cloud giants and GPU manufacturers joining forces to bind head model companies.
- Taiwan's "TryAI" government AI application platform expands, supporting 32 models and a "trial before procurement" mechanism
The Ministry of Digital Development disclosed that the "Government AI Application Platform" has integrated 32 language models such as TAIDE, Gemini, and ChatGPT and 20 general AI tools to help agencies verify the results before purchasing through a safe and compliant trial environment, greatly lowering the threshold for government scenario introduction.
- The "Fengshun" AI climate prediction system of the China Meteorological Administration ranks among the best in international competitions
The "Fengshun" AI sub-seasonal-seasonal climate prediction system developed by the China Meteorological Administration and a number of scientific research institutions temporarily ranked among the top in the competition hosted by the World Meteorological Organization and the European Center for Medium-Range Weather Forecasts, demonstrating that the AI-based global climate anomaly prediction ability has reached a commercialized level.
- The model parallel training system of China University of Science and Technology won the only best paper award at ICCD 2025
The large model parallel training system cooperated by the University of Science and Technology of China and the National University of Singapore won the only best paper award at ICCD 2025, indicating that China has internationally leading achievements in large model system-level optimization and computing power utilization efficiency.
- Intuit and OpenAI have reached a multi-year cooperation of more than $100 million to introduce financial and tax capabilities into ChatGPT The
two sides announced the signing of a multi-year cooperation agreement with an amount of more than $100 million, and in the future, users can directly access Intuit's finance, taxation and enterprise service capabilities in ChatGPT, forming a demonstration effect on the commercialization prospects of "AI assistant + vertical financial scenarios".
- Google launches Google Scholar Labs to answer scientific research questions with generative AI
Scholar Labs is launched as an experimental feature, trying to use generative AI to provide researchers with more structured and traceable literature reviews and question answers, which is regarded as a key pilot for large models to in-depth academic search and knowledge services.
- Musk's xAI was revealed to be negotiating $15 billion in financing, with a valuation of $230 billion
According to multiple reports, xAI is in in-depth negotiations with investors on a new round of financing of up to $15 billion, and the valuation is expected to increase to about $230 billion after the completion of this round, once again highlighting capital's long-term bets on general models and infrastructure tracks.
- The Information Policy Council announced the top ten key technologies in 2026, and physical AI and industry large models were selected
The Software Institute of the Information Policy Council released the list of the top ten key technologies in 2026 at the STI TECH DAY, which included physical AI in the real world, generative AI models in specific fields and responsible AI into the key directions, indicating the technical trend of regional industrial policies and enterprise R&D layout in the coming year.
Frequently Asked Questions (Q&A)
Q: What is the most iconic AI news piece globally today?
A: Google released and immediately launched Gemini 3 in core products such as Search and Gemini, which is not only a single model upgrade, but also pushes "AI-first search" and multi-modal interaction to the default experience, which has a profound impact on the entire search and information distribution ecosystem.
Q: Which dynamic in China has the strongest short-term perception of ordinary people?
A: Nanjing's generative AI literacy white paper for primary school students and Taiwan's TryAI government platform, one for families and campuses to use AI "how to use it safely", and the other to make government services more efficient and transparent, will be gradually perceived in daily life scenarios.
Q: From the trend point of view, what are the two main lines revealed by today's news?
A: First, "computing power + model" is still frantically attracting money, and the huge financing from Anthropic to xAI shows that the competition at the basic layer is far from peaking; Second, "governance + literacy" has accelerated significantly, and government platforms, education white papers, and industry norms have jointly pointed to a more controllable and explainable AI implementation.
Q: How does Google Gemini 3 differ from other large models at home and abroad?
A: Gemini 3 emphasizes multi-modal long-chain inference and deep integration within the product, directly entering search and office scenarios; Many domestic and vertical models pay more attention to industry adaptation, Chinese scenarios and localized deployment, and the two form a dislocation competition between "general intelligence" and "industry deep cultivation".
