THE NEWSROOM / AI 新闻
Source-linked briefings on models, agents, products and practical AI.
YYSI publication dates appear on the left; original announcement dates appear under each brief.
Search and categories filter this page. Use the page links below for earlier news.
Product updates OpenAI
According to The Verge, OpenAI is rolling out its Images 2.6 model to all ChatGPT and Codex users on web, desktop and mobile, adding native GIF creation from prompts or uploaded photos, per The Verge. Users can adjust animation speed, edit individual frames and share results. OpenAI also launched GPT-Image-2.6 Flare and Sunburst in its API and says users now generate over 4 billion images weekly.
Practical guides AI industry
According to The Verge, In the 2026 US midterms, voters in dozens of local ballot measures approved restrictions on data center development, with 33 of 45 measures limiting projects as of Wednesday, according to a Redbud Consulting tally cited by The Verge. NBC News exit polling found 76 percent of voters opposed data centers in their communities, and several winning candidates back moratoriums.
Product updates AI industry
According to Cloudflare Blog, Cloudflare launched analytics that classify personal AI agents as a distinct traffic category, showing which agents visit a site, what paths they request and which accounts they act for. Site owners can now set per-agent access policies — allow, block, or block only on ad-supported pages — under a renamed AI Crawler and Agent access policies setting. Account-level agent visibility is in early access.
Models & research AI industry
According to IT Home, At the World Internet Conference AI forum, Moxin Technology released MoWorld 2.0, a 4D world model trained and deployed entirely on Huawei Ascend compute. The company says it models physics such as gravity, elasticity and fluids to output reusable, real-time-renderable 4D Gaussian assets, with video inference rising from about 50 to 70 fps. Moxin claims inference costs as low as 30% of comparable GPU setups and is testing uses in embodied AI, autonomous driving and digital twins.
Models & research AI industry
According to QbitAI, China Merchants Group's Lion Rock AI Lab released Lion-ER-35B-A3B, a Mixture-of-Experts vision-language model for embodied reasoning with about 35B total and 3B active parameters. The lab says it improved on its base model across ten embodied benchmarks, scoring 65.29 on an equal-weight average, using a capability-driven data engine rather than larger datasets. Weights are open on Hugging Face.
Product updates AI industry
According to IT Home, AMD signed a memorandum of understanding in Seoul with AI NPU maker FuriosaAI and DPU vendor MangoBoost to build an open heterogeneous AI computing ecosystem. The three plan a reference cluster in Seoul next spring combining AMD CPUs and GPUs, FuriosaAI's RNGD NPU and MangoBoost's BoostX DPU to match workloads to hardware. MangoBoost also plans to use AMD Instinct accelerators in its AI factory.
Product updates Google
According to IT Home, Google said it plans to invest $15 billion in Malaysia to expand data centers and cloud computing, with construction in 2027 and 2028. The company will build new computing facilities and fund water security, power supply and local skills training. Google estimates the build phase will add $1.1 billion to Malaysia's economy and support over 37,000 local jobs annually.
Models & research AI industry
According to IT Home, Tencent released the official version of its Hunyuan Hy4 large language model, saying it significantly expanded pre- and post-training and added vision and computer-operation capabilities for the first time. Tencent claims Hy4 joins the strongest open-source models, though it acknowledges a large gap to top closed models. Hy4 is live on Tencent Cloud and OpenRouter at the same API price as the preview, with weights and a system card planned for open release.
Product updates AI industry
According to The Verge, Nexus Mods, the largest PC gaming mod repository, will require authors of newly uploaded mods to declare whether AI was used, with AI-assisted mods needing a public explanation of how and where. AI tags will be more prominent, and users get controls to hide AI content. The site declined an outright ban, citing AI's spread across creation, and says it won't use or license mods for AI training.
Models & research Anthropic
According to TechCrunch, Anthropic researchers published an open source technical specification and partial code for FlashProve, a system for "provable inference" that cryptographically verifies which model produced a given output. It relies on labs publishing model fingerprints that do not reveal weights, and is designed to run on local devices. The work is intermediate research, with no deployment planned at Anthropic or elsewhere.
Product updates OpenAI
According to IT Home, OpenAI consumer product lead Adam Fry said in an interview with Big Technology that ChatGPT now has 1.2 billion weekly active users, up from 1 billion in the past 60 days, roughly 100 million new users per month. He also said ChatGPT handles about one-fifth of Google's search query volume, generates 1.5 billion images weekly, and has 250 million weekly users of voice input and voice conversation combined.
Product updates AI industry
According to The Verge, Apple is letting developers integrate Siri AI's new capabilities across third-party apps through an iOS 27.3 Developer Preview, The Verge reports. The assistant can use personal context, natural-language prompts and on-screen awareness, and a new "app skills" feature lets developers bundle tasks into single actions. Apple is also extending the expanded capabilities to its own apps, including Mail, Maps, Messages and Photos.
Product updates AI industry
According to TechCrunch, Fal, a San Francisco startup that hosts and runs AI video, image and audio models, raised $1.3 billion in a round led by Bessemer Venture Partners, with Nvidia, Sound Ventures and T. Rowe Price participating, Bloomberg reported. The deal values Fal at $16 billion, double its April valuation. Fal announced Series E and F rounds together, totaling $1.65 billion, and surpassed $800 million in annualized revenue earlier this year.
Product updates AI industry
According to TechCrunch, Netflix announced a new AI Discover tab that recommends shows and movies based on users' mood, available time, viewing preferences and trending titles, using a new intelligence layer that responds to real-time input and keeps learning. The tab is rolling out in the U.S. to English-language users on supported TVs, with wider availability planned in coming weeks. Netflix calls it a test kitchen for discovery features.
Product updates AI industry
According to IT Home, Apple has begun testing Apple Intelligence in China with the first iOS 27.3 developer beta, with a public rollout planned for December. The first features include live translation, writing tools, visual intelligence, photo editing and natural-language search, notification summaries and a more conversational Siri. Siri can also draw on Alibaba's Qwen model for knowledge queries and content creation, per IT之家.
Product updates Google
According to IT Home, Google DeepMind announced cheminformatics skills for Google Antigravity that combine LLM planning and reasoning with deterministic chemistry tools such as RDKit and public databases ChEMBL and PubChem. In a lung-cancer demo, an agent screened 568 compounds in about an hour, analyzed matched molecular pairs and proposed synthesizable macrocyclic candidates. Google says outputs are testable hypotheses, not drugs, and must be validated experimentally. Access is open to Antigravity users, with free high-usage tiers for up to 10,000 scientists.
Product updates Google
According to TechCrunch, Google said it will disable AI tools for K-12 students next year unless school administrators explicitly enable them, revising its approach across Gemini, NotebookLM, Gemini in Classroom and Search. Because AI features are now embedded in Search, schools that restrict AI entirely would lose Google Search access. Google announced the change alongside the school launch of Guided Learning, its AI tutor-style tool.
Practical guides AI industry
According to TechCrunch, Apple told developers it will require all apps offering dating or romantic-relationship features, including AI companion apps, to carry an 18+ App Store rating. From November 16, developers must add new content descriptors so age ratings reflect app features. Apps sharing users' real-world locations with others must also disclose this. The policy extends Apple's broader child-safety and age-assurance efforts to AI companions.
Practical guides Google
According to IT Home, An advocate general at the EU Court of Justice said Google's Gemini chatbot may infringe news publishers' copyright in specific cases, such as bypassing paywalls or reproducing identifiable expression from articles. He added that training LLMs on protected material is allowed under the text and data mining exception, and that bare facts are not protected. The opinion is non-binding but often followed.
Product updates Google
Google said Guided Learning, its step-by-step tutoring feature that coaches students instead of giving direct answers, is now globally available in Google Classroom and is being built into Search for student accounts, with a pilot in select U.S. schools later this month. Google also changed school controls so Gemini, Gemini Notebook, Gemini in Classroom and Search are off for K-12 student accounts until administrators enable them.
Try another keyword or category. Saved tools stay in this browser only.
