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 Microsoft
Microsoft released version 1.0 of its Foundry extension bundle for Azure Developer CLI, adding agent setup, local testing via Agent Inspector, and deployment of hosted agents on managed compute. Configuration lives in azure.yaml alongside code, and the release adds evaluation tooling plus a Foundry Skill for coding agents. It requires azd 1.37.0 or later and incurs Azure charges.
Product updates Google
Google says its Vids video creation tool now runs on Omni 1.2 Flash, its newest video generation model. Anyone with a Google account or Workspace subscription can generate up to three 15-second videos per month at no cost. Google claims improved speech narration and consistent characters and scenes when using uploaded reference images. The model is available through vids.new.
Product updates Meta
According to TechCrunch, Meta said its Muse Voice feature, which lets US users call and video chat with its consumer AI agent Muse, is now available to everyone in the country. The feature, first shown at Meta's September Connect conference, relies on new Muse Realtime Avatar and Muse Realtime Voice models that drive an animated avatar with matching expressions and gestures. Meta also plans a Muse wearable before year-end.
Product updates AI industry
According to The Verge, Meta CEO Mark Zuckerberg announced that the Muse AI agent now supports real-time voice conversations, with an on-screen avatar that animates in sync with its speech. Users can adjust voice tone by asking the agent or choosing from settings. The feature is rolling out in the US on the app, with Meta saying it will arrive soon on its AI glasses and the forthcoming Muse Charm gadget.
Product updates Anthropic
According to TechCrunch, Anthropic announced a Claude tool called Personal Finance that lets users link bank and credit card accounts through Plaid and ask questions about balances, spending and fees. Access is read-only, and chats in the hub are not used for training. It begins rolling out in beta this week to Pro, Max and Ultra subscribers in the U.S., per TechCrunch.
Models & research Google
Google DeepMind launched Gemini Omni 1.2 Flash, a video generation model now generally available on the Gemini API and Gemini for business. It extends native clip length to 15 seconds, adds multi-turn conversational editing and stronger reference adherence, and cuts the price 30% to $0.07 per second for 720p video. Google also cites more natural speech and lip-sync.
Models & research OpenAI
According to The Verge, OpenAI is launching GPT-6 Cyber, a cybersecurity model it says finds vulnerabilities more effectively while going out of bounds less than 1 percent of the time in testing, down from about 5 percent for GPT-5.6 Cyber. It scored 48.5 percent on the ExploitGym benchmark and is classified Critical under OpenAI's Preparedness Framework. Access is rolling out to Daybreak Red defensive-security customers.
Models & research AI industry
According to TechCrunch, Axiom Math, a $1.6B startup, said its fleet of AI agents translated the roughly 7,654-page Stacks Project into about 16 million lines of Lean code, formally verifying one of the largest human-written bodies of mathematics. The company trained its own math-focused models and let agents pick parallel tasks. It argues verified results give AI agents a reusable foundation for code and science.
Product updates AI industry
According to TechCrunch, Dropbox announced a $350 million acquisition of cloud storage company Tigris Data, which it says is its largest deal to date, to strengthen AI infrastructure and support DBX Agent Cloud, a storage platform for businesses and developers building AI models and agents. The deal is expected to close in Q4 2026. Dropbox also detailed AI features for its core product, including model choice and connections to ChatGPT, Claude and Muse.
Product updates NVIDIA
According to TechCrunch, At its developer conference, Box unveiled Box Content Retrieval, which embeds Nvidia's open-weight Nemotron embedding models into its core product so AI agents can retrieve relevant document chunks instead of reading whole files. Nvidia's Kari Briski said queries could be answered four times faster with 20% fewer errors and 90% fewer tokens. The feature rolls out early next year to Box Enterprise Advance customers.
Product updates AI industry
According to IT Home, DJI released the Matrice 5T, a light industry drone priced at 57,488 yuan in mainland China. It is DJI's first light industry drone to include LiDAR, millimeter-wave radar and full-color omnidirectional vision as standard. It carries thermal and multi-focal cameras, flies up to 59 minutes, and its onboard AI recognizes over 20 target types, with open compute for third-party models.
Product updates AI industry
According to IT Home, SAIC Motor's board approved a plan to form a joint venture with Momenta and Yandian Technology to strengthen advanced driver-assist R&D, according to an IT Home report on the company's announcement. The venture would have 300 million yuan in registered capital, with Momenta holding 51%, SAIC 25% and Yandian 24%. Momenta, listed in Hong Kong, positions itself as a physical-AI foundation model provider.
Practical guides OpenAI
According to IT Home, Amid nationwide pushback over data-center power, water and noise, OpenAI published a National Community Compact and Compute Campus Policy Blueprint. The company says it will cover its own electricity costs and grid-expansion expenses, provide financial guarantees before major spending, disclose site power and water use, avoid broad NDAs with officials, and hold community consultations. OpenAI's Chris Lehane said a southern Ohio site is expected to create 2,500 long-term jobs.
Product updates Meta
According to IT Home, Meta has acquired Conductor, a 2024-founded startup whose AI coding tool lets users run multiple AI agents in parallel workspaces to advance separate software projects. Conductor CEO Charlie Holtz announced the move, saying he will work with Meta AI chief Alexandr Wang and Superintelligence Labs product lead Nat Friedman. Meta confirmed the deal to TechCrunch without disclosing terms. Conductor's cloud service closes November 20, with stored data to be deleted.
Product updates Anthropic
According to IT Home, Media reports that Cat Wu, who led product for Anthropic's Claude Code and helped take it from an internal tool to a commercial product, has left Anthropic and joined OpenAI as head of product management for Core Products. She will work on agents and core products alongside Nick Turley and Ashley Alexander, reporting to Tibo Sottiaux. OpenAI president Greg Brockman welcomed the move.
Agents & tools Kimi
According to IT Home, Kimi announced Kimi Hosted Agents, offering its agent Harness as a cloud service so enterprises and developers can run agents without building their own infrastructure. It provides 24/7 hosted environments for long-running tasks and multi-agent collaboration, isolated sandboxes, network access controls, audit logs and usage billing. It ships with 140+ Skills, 30+ plugins and eight financial agent templates, with pricing based on sandbox hours, inference and plugin calls.
Models & research AI industry
According to IT Home, The University of Science and Technology of China released structRFM, an open-source RNA sequence foundation model that incorporates secondary-structure knowledge into large-scale pretraining. Trained on about 21 million computationally annotated sequence-structure pairs, it improves secondary and tertiary structure prediction and functional inference tasks such as splice-site prediction. The team published the work in Nature and released the paired data, model weights and source code.
Models & research OpenAI
According to IT Home, OpenAI reported that an unreleased cybersecurity model, run through an automated discovery, validation, exploitation and patching pipeline, surfaced 2,308 candidate Linux kernel issues and reproduced 1,724 of them. It produced 948 distinct local privilege-escalation exploits, up from 166 in a July test with GPT-5.6 Sol, at about $1.21M in API cost. OpenAI says 61 flaws were already fixed upstream.
Product updates AI industry
According to TechCrunch, On Uber's third-quarter earnings call, executives said AI agents placed just four rides and about 500 Uber Eats orders during the week of Halloween, out of roughly 320 million total trips and orders. CEO Dara Khosrowshahi called agent traction "very, very early and very limited" but said Uber expects agents to add demand over time. The figures echo DoorDash's report of only a few hundred agent-placed orders.
Product updates NVIDIA
NVIDIA released CUDA Toolkit 13.5 to general availability, moving Vera Rubin GPU support from early developer access to production use. The release extends Windows on Arm support to DGX Station for Windows and adds locality domain controls for SM partitioning, MPS v3 scheduling features including weighted timeslicing and SM borrowing, CUDA Tile multi-GPU communication over NVLink/NVSwitch, and cuda.core 1.2 batched memory copies.
Try another keyword or category. Saved tools stay in this browser only.
