Discover the AI Tools

Baidu's Shen Dou reveals the company's unique approach to AI adoption: instead of forcing employees to use AI tools, they give each worker a monthly allowance of 1,000 yuan to freely explore mainstream large models. The strategy aims to lower barriers and let employees discover AI's value naturally. Meanwhile, Baidu's AI revenue has surpassed half of total revenue for the first time, and founder Robin Li proposes a new metric for the AI era: Daily Active Agents (DAA).

At WAIC 2026, StepZen showed off the STEPX Neo, a prototype smartphone designed from the ground up for AI agents. Unlike traditional phones, it runs on the Step AOS system with an AI assistant named Amoo that can handle tasks like booking hotels or finding concert tickets. It's still a prototype, but it hints at a future where phones become smart agent platforms.

At the 2026 World AI Conference, WeRide introduced WIIT, a foundation model that helps AI grasp real-world dynamics by breaking down environments into basic 'physical facts.' This could be a game-changer for autonomous driving and robotics, moving AI beyond data processing to genuine spatial and behavioral understanding.

Autonomous driving company WeRide has launched WITT, a physical AI large model that introduces the concept of 'minimum physical fact unit' to help AI better understand real-world scenes. By breaking down complex multimodal information into core facts, WITT aims to make autonomous vehicles smarter and safer. This move marks a significant step toward more human-like AI perception and reasoning in transportation.

拥有千万粉丝的内容创作者“影视飓风”推出了一门49元的AI实战课程，上线首日即售出超10万份，单日销售额突破490万元。这门课程涵盖AI图像生成、视频制作等基础应用，以及模拟机械臂运镜、子弹时间特效等进阶技巧。然而，高销量背后用户评价两极分化，部分用户反映课程内容模板化、与宣传存在差距。

Tongyi Lab's Wan-Streamer v0.2 slashes response delay to just 550 milliseconds, enabling truly natural, face-to-face conversations with AI. This end-to-end multimodal model integrates listening, seeing, speaking, and expression in a single Transformer, outperforming traditional cascading systems. With improved video resolution and full-body digital humans, it's set to transform applications from language practice to psychological counseling.

At the 2026 World AI Conference, Wang He, founder of Galaxy General Robotics, predicted that embodied intelligence will hit a breakthrough before 2028, matching ChatGPT's conversational prowess. He envisions foundation models achieving 70-80% success on untrained tasks, powered by massive data and efficient training. The talk sparked excitement about robots handling chores and complex decisions, hinting at a transformative leap for AI.

Chongqing has officially launched 'Yu Xiaojian', an AI medical assistant that helps residents with intelligent triage, cloud companionship, and report interpretation. Developed by the city's health commission, big data bureau, and Ant Group, the service now covers 233 public hospitals, serving over 2 million users since its pilot phase.

At WAIC 2026, DeWu app teamed up with Qianjue Robot to launch the world's first AI authentication robot. This bot uses DeWu's massive product database and an 'AI brain' to spot fake items with over 99.9999% accuracy. It can scan a pair of sneakers and generate a report in about 5 seconds, covering everything from shoes to cosmetics. The move signals how AI and robotics are moving into quality inspection, making authentication faster and smarter.

NVIDIA just dropped the Nemotron-3 Embed series, a family of embedding models built for real-world RAG, code search, and agent memory. The 8B version snagged the top spot on the RTEB benchmark, beating all other open-source models. There's also a 1B lightweight version and a 4-bit variant optimized for Blackwell hardware—perfect for cost-conscious setups.

Google has officially renamed NotebookLM to Gemini Notebook, integrating it into the Gemini brand while adding cloud computing capabilities. Users can now write and run code directly within the notebook, enabling complex data analysis without leaving the app. The update rolls out first to AI Ultra and enterprise users, with broader availability soon.

Google Vids just got a major update: upload a selfie and a voice recording, and it creates a digital avatar that looks and sounds like you. The new Gemini Omni model powers conversational video editing, letting you tweak backgrounds, lighting, and effects step by step without starting over. This turns Vids into a serious competitor to AI video startups like HeyGen and Synthesia.

AI旅行平台Fora完成6000万美元D轮融资，估值达到10亿美元，正式跻身独角兽行列。本轮由Forerunner和Tactile Ventures领投，Insight Partners和Thrive Capital等参投，累计融资达1.385亿美元。Fora计划利用新资金扩大AI助手Via的应用，帮助旅行顾问处理行政事务，提升效率。

DoorDash has launched a limited beta of dd-cli, an AI command-line tool that lets developers order food directly through an AI assistant. Currently available to macOS developers in the US and Canada via waitlist, the tool represents a step toward agentic commerce, where AI agents handle tasks like ordering, grocery shopping, and finding deals without opening an app.

1Password与AI助手Claude达成合作，推出了一项看似矛盾却巧妙的功能：让AI帮你填写密码并执行任务，但真正的密码被锁在保险箱里，AI根本接触不到。当Claude想登录网站时，1Password会明确告知要使用哪个凭证并征求你的同意，授权仅在当前任务有效，完成后立即撤销。目前该集成仅支持Mac平台，需要同时安装1Password桌面版和浏览器扩展，以及Claude桌面版和Chrome扩展。

A hacker leaked Suno's internal source code, revealing how the AI music platform scraped millions of songs from YouTube, Deezer, and Genius to train its models. The breach, which occurred in late 2025, also exposed user data including emails and partial credit card info. This could intensify copyright lawsuits from major record labels.

Several Chinese banks have launched AI-themed credit and debit cards that offer large model tokens and AI membership benefits instead of traditional points. The move targets tech-savvy users and developers, but experts question whether ordinary consumers will find the perks appealing. The cards, developed with UnionPay and AI firms like Kimi and Alibaba Cloud, reflect a growing trend of integrating AI into financial products.

xAI's Grok just launched an automation feature that lets you set up tasks to run on a schedule, trigger on incoming emails, or even reply on your behalf. It's like having a personal assistant that never sleeps—create a task in plain English, choose when it runs, and let Grok handle the rest. Available now on grok.com and the Grok app.

Google is testing a new feature for Gemini that lets users tweak the AI's voice with four sliders: speed, energy, formality, and friendliness. Discovered in a beta version of the Google app, this update aims to make interactions more natural and personalized, moving beyond preset voice options. It's a step toward more human-like AI conversations.

Zhipu AI's annual recurring revenue (ARR) has reportedly reached $1 billion as of July 2026, growing 15-fold in just five months. The company's early focus on AI coding capabilities, with rapid model iterations and competitive pricing, has driven this explosive growth. This milestone highlights the booming commercialization of generative AI, especially in coding and video generation, while intensifying competition among domestic and overseas players.