中文 ↗RSS ↗

THE NEWSROOM / AI 新闻

AI News · Page 5

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.

SOURCE

Product updates Amazon / AWS

Amazon Bedrock Managed Knowledge Base adds single-shot RAG and retrieval-only modes

AWS says Amazon Bedrock Managed Knowledge Base now supports single-shot RAG and retrieval-only requests through AgenticRetrieveStream, alongside existing agentic retrieval. A maxAgentIteration setting lets developers pick retrieval-only, single-shot RAG or multi-search agentic retrieval per request, aiming to cut latency and model calls for simple lookups while keeping planning for complex questions.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Models & research Amazon / AWS

Anthropic's Claude Fable 5.2 and Mythos 5.2 launch on Amazon Bedrock

AWS announced that Anthropic's Claude Fable 5.2, described as its most capable generally available model for coding, math and science, is now available on Amazon Bedrock and Claude Platform on AWS. Claude Mythos 5.2 is offered only through Anthropic's verification programs for cyber security and life sciences. AWS also detailed data retention rules and Enterprise Frontier Safeguards for the covered model.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Product updates Microsoft

Microsoft previews agentic mode in Azure Content Understanding

Microsoft announced a preview of agentic mode in Azure Content Understanding, part of Foundry Tools. Unlike the default single-pass standard mode, agentic mode wraps the language model in a document-native harness with a reasoning loop and tools for navigating text and tables, inspecting visual evidence, calculating and validating results. Microsoft positions the two modes as points on a price-performance curve, recommending standard mode first and agentic mode for harder multi-step tasks.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Product updates Amazon / AWS

NVIDIA Nemotron 3.5 Super VL arrives in Amazon SageMaker JumpStart

AWS announced that NVIDIA's Nemotron 3.5 Super VL vision-language model is available in Amazon SageMaker JumpStart. The 120B-parameter hybrid Mixture-of-Experts model has 12B active parameters, a C-RADIO vision encoder, up to 1M-token context and a built-in speculative decoding head. JumpStart offers NVFP4 and BF16 variants, targeting document intelligence, GUI understanding for computer-use agents, video QA and multimodal agents.

AI summary · Official sourceOriginal source ↗Read brief →
REPORTED

Product updates AI industry

Apple Sells $49.99 Intelligence Usage Packs for Cloud AI Features in Logic Pro

According to The Verge, Apple has added a "Buy More Usage" option to Logic Pro priced at $49.99, letting users extend access to Apple Intelligence features that rely on cloud models from OpenAI or Google Gemini. The packs are tied to the purchasing Apple Account, unused usage rolls over monthly and expires after 12 months. Apple reportedly plans to extend the option to other Creator Studio apps.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Models & research Anthropic

Anthropic Releases Claude Fable 5.2 and Limits Claude Mythos 5.2 to Vetted Organizations

According to The Verge, Anthropic announced Claude Fable 5.2 and Claude Mythos 5.2, which it says improve coding, reasoning and vision and generate output faster than the 5.1 versions. Fable 5.2 is available today and can orchestrate agent teams and long-running projects; Mythos 5.2 is initially limited to vetted organizations. Anthropic claims prerelease testing included solving two open black-hole physics problems and formalizing Fermat's Last Theorem in 30 hours.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Models & research Meta

Meta Releases Muse Ember, a Lightweight Model for Agent Execution

According to TechCrunch, Meta Superintelligence Lab released Muse Ember, its lightest and most cost-efficient model, aimed at high-volume agent work such as tool calls, coding subtasks, routing, extraction and long document parsing. Meta claims it is cheaper and smarter than Anthropic's Haiku and OpenAI's Luna, at 10 cents per million input tokens, and says it pairs with the larger Muse Wonder model in agentic setups.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates Anthropic

Apiiro Finds Thousands of Secrets Leaked via Claude Code Settings Files on GitHub

According to Ars Technica, Security researchers at Apiiro report that Claude Code's local settings file, which can store approved shell commands verbatim, has been uploaded to public GitHub repositories over 100,000 times, with roughly 1,100 files from 63 organizations containing secrets such as API tokens, database passwords and private keys. Anthropic's Boris Cherny said the team is looking into prevention and may add secret scanning.

AI summary · Media reportOriginal source ↗Read brief →
SOURCE

Product updates Google

Google Maps Platform integration arrives in Google AI Studio

Google announced that developers can now use Google Maps Platform directly inside Google AI Studio, so they can prompt Gemini with real-world concepts and get a live interactive preview without leaving the workspace to configure billing and credentials. The free tier requires no credit card and is limited by daily quotas; production use requires a paid Cloud billing account.

AI summary · Official sourceOriginal source ↗Read brief →
REPORTED

Product updates AI industry

Disney and Adobe Partner to Build Custom Generative AI Models for Production

According to The Verge, The Walt Disney Company and Adobe announced a multiyear partnership at Adobe's MAX conference to develop proprietary, artist-led generative AI models for Disney's production pipeline. Under Adobe Firefly Foundry, models trained on Disney character attributes, environments and visual styles will generate commercially safe outputs. Disney gets early access to Adobe technology across theme parks, cruise line and ESPN content, building on Imagineering's existing Firefly use.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Models & research AI industry

Meituan LongCat Open-Sources LongCat-2.5, a 1.6T-Parameter Multimodal Model With Million-Token Context

According to IT Home, Meituan's LongCat team has open-sourced LongCat-2.5 under the MIT license, allowing commercial inference and further training without extra authorization. The natively multimodal model targets long-horizon agent tasks, has 1.6 trillion total parameters with about 48 billion activated per token, and supports a one-million-token context. It uses discrete visual tokens processed by the main model and a W4A4 INT4 inference scheme, with optimizations for domestic AI chips.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates Meta

Meta to Expand AI Age Estimation Globally to Protect Teens

According to IT Home, Meta announced it will roll out AI age-estimation technology worldwide, extending Instagram's Teen Accounts protections to millions more teens even when their stated birth dates indicate they are adults. The company is also using Apple-provided age-range signals in Australia to help verify users' ages, with the approach expanding to Facebook and later Threads. The move follows lawsuits and an August settlement with US states over alleged harm to young users.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Practical guides AI industry

South Korea Bans AI Glasses and All Electronics From 2026 College Entrance Exam

According to IT Home, South Korea's education ministry issued exam-day guidance barring phones, tablets, AI glasses, smartwatches and Bluetooth earphones from the national college entrance exam on Nov 19. Each test site will station a staffer to check for AI glasses; wearing them, even unused, voids this year's score and next year's eligibility. The ministry cited multiple AI-glasses cheating cases and plans signal detectors next year.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Practical guides AI industry

Norway Halts Grid Capacity Reservations for New Data Centers Pending Licensing Rules

According to IT Home, Norway's grid operator Statnett will temporarily stop reserving grid connection capacity for new data centers until the government creates a licensing regime, Reuters reported via IT之家. Energy minister Terje Aasland said Statnett has received 26,000 MW of new capacity requests this year, exceeding national generation, with data centers at 75%. New rules would prioritize defense, emergency, health and critical infrastructure projects.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Models & research AI industry

ByteDance Releases Seedream 5.5 Image Model With 100-Round Editing and Native Transparency

According to IT Home, ByteDance's Seed team released Seedream 5.5, an image creation model adding pixel-level preservation across up to 100 rounds of edits, native RGBA transparent output, and improved text rendering. The company says its severe text-corruption rate fell to 0.4%. It is rolling out on Jimeng, Doubao and Volcano Ark, with API cost around 0.28 yuan per 2K image edit, under half of Seedream 5.0 Pro. A Pro version is planned.

AI summary · Media reportOriginal source ↗Read brief →
SOURCE

Product updates NVIDIA

NVIDIA announces investment in Point2 and plans for Motif, RLWRLD and Upstage

At the MSIT-NVIDIA AI Startup Partnership Forum in Seoul, NVIDIA said it has invested in Korean AI interconnect startup Point2 and plans further investments in Motif Technologies, RLWRLD and Upstage. The startups span foundation models, physical AI and AI infrastructure. NVIDIA also cited more than 1,000 Korea-based members of its Inception startup program. Investment amounts were not disclosed.

AI summary · Official sourceOriginal source ↗Read brief →
REPORTED

Product updates AI industry

China Telecom Unit and UBTech Sign Strategic Pact to Scale Bionic Humanoid Robots

According to IT Home, Tianyi Telecom Terminal, a wholly owned China Telecom subsidiary, signed a comprehensive strategic cooperation agreement with UBTech on November 9. The partners plan to push household companion robots past one million units and bionic and wheeled humanoid robots past 100,000 units over three years. They also unveiled a joint robotics innovation lab and will pair UBTech robot capabilities with telecom cloud, AI model, quantum, satellite, security and eSIM services.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Practical guides AI industry

China's Cyberspace Administration Launches Unified Filing System for Generative AI, Algorithms and Blockchain

According to IT Home, China's Cyberspace Administration launched a unified online filing system on November 9, 2026, consolidating registration for generative AI, algorithms, deep synthesis and blockchain services. The new portal replaces the separate algorithm and blockchain filing systems, which stopped operating the same day. Companies and the public can search filings by entity name, filing name or filing number.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates NVIDIA

NVIDIA AI Day in Seoul: Sovereign AI Deals with Korean Government, Samsung, LG and Hyundai

According to IT Home, At NVIDIA AI Day in Seoul, NVIDIA announced cooperation with South Korea's MSIT on a sovereign AI ecosystem spanning AI chips, data centers, talent, startups, models, agents and security. Samsung is expanding collaboration on accelerated computing and AI for semiconductor R&D, while LG Electronics demonstrated Isaac GR00T-driven robot manipulation with its CLOiD robot in a manufacturing setting. Kakao Mobility and Hyundai Motor Group also outlined work with NVIDIA platforms.

AI summary · Media reportOriginal source ↗Read brief →