Discover the AI Tools

Ant Group has open-sourced Ling-3.0-flash-VL, a 124B-parameter multimodal model that natively handles text, images, and video. With a 256K context window and a clever visual feedback loop, it not only expands AI's capabilities but also boosts text intelligence. This release challenges the notion that adding vision dilutes language skills.

曹操出行与豆包合作推出的AI打车服务，已在北京、杭州、苏州上线。用户只需在豆包中咨询路线、美食或娱乐信息，系统就能自动生成出行方案并直接叫车。未来，双方还计划拓展智能控车等功能，并探索Robotaxi。这一创新让出行变得更智能、更便捷。

在京东全球科技探索者大会上，京东技术委员会主席曹鹏宣布启动“京东物理AI加速计划”，聚焦数据、机器、制造、销售、物流和服务六大领域。计划包括构建全球最大具身数据训练中心、机器人制造基地，并投入超百亿元推动机器人进入千家万户。京东的目标是打造全球最大的实体世界协作系统，成为领先的物理世界运营中心。

Samsung is teaming up with Mistral AI, a French startup founded by ex-DeepMind and Meta engineers, to bring advanced AI into its semiconductor business. The goal? Faster chip development, better manufacturing precision, and more stable yields. This move is part of Samsung's broader push to weave AI into every layer of its chip operations, with other partnerships like OpenAI, Arm, and Qualcomm already in the mix.

China Unicom and Meizu have teamed up to launch the 'Xiaomo Fang,' a pocket-sized AI phone weighing just 99 grams. With a 4-inch square screen and a focus on voice control, it can order coffee, identify objects, and even translate calls in real time. The device also offers cloud storage and smart calling features, making it a unique entry in the AI-native terminal space.

DeepSeek, the AI unicorn, has officially engaged CITIC Securities to prepare for a Shanghai STAR Market IPO, aiming to file this year and list next year. The company's revenue surged to 475 million yuan in the first seven months of 2026, about ten times its 2025 total. A new funding round targets a 500 billion yuan valuation, following a June round that raised $7.4 billion. However, off-market SPV trading has raised governance concerns, prompting founder Liang Wenfeng to personally vet investors.

Jacob Coxon, a senior researcher at Anthropic, has resigned, warning that the AI industry's race toward superintelligence is dangerously reckless. He cites concerns about recursive self-improvement and the lack of government oversight. His departure highlights growing unease among researchers about the pace of AI development and its potential risks to humanity.

OpenAI's South Korea head, Harrison Kim, revealed that the company is working closely with Samsung Electronics on next-generation chip development. Samsung, a major ChatGPT user, is also a key partner in memory chip research. As AI demands faster and more complex computations, the need for advanced memory chips grows, making Samsung a priority partner for OpenAI's local operations.

Meta has unveiled Muse, a personal AI agent that runs on cloud-based virtual machines and can handle tasks like booking tickets or filling forms. It's available in the US for users 18 and older, with plans to expand. Zuckerberg assures it won't go rogue, and it's designed to be simple and helpful.

2026年外滩大会以“创造AI新经济”为主题，展示了AI从概念走向实际应用的全景。从能工作的机器人到无处不在的智能体，AI正在改变工业、生活和医疗。大会不仅呈现了技术的进步，更探讨了AI如何惠及普通人，以及背后的安全与人文思考。

Google has opened its Gemini Daily Briefing feature to all free users, no longer requiring a paid subscription. This smart assistant compiles your tasks, emails, and key info into a neat morning summary. Available on web and mobile, it's rolling out in the US first, with audio playback on the horizon. Here's what you need to know.

The 2026 Bund Conference in Shanghai debuts an AI Arts Festival, bringing AI-generated art, films, and music to a physical venue. From a 1985 AI painting to a live concert with a virtual idol, the event explores the evolving relationship between humans and machines in the creative realm.

9月9日，外滩大会上的AI编程挑战赛创下吉尼斯世界纪录，成为国内首个以AI编程为主题的纪录挑战。近七成参赛者未满18岁，最小的只有6岁，最终9岁的罗安歌凭借AI制作的陶瓷主题数字博物馆夺冠。这场比赛不仅展示了AI如何降低创作门槛，还催生了国内首个多平台AI编程技术规范。

OpenAI just dropped ChatGPT Images 2.5, and it's a big deal for anyone who loves generating images. The new model halves the time you wait for results, sharpens details, and makes lighting look more natural. But the real showstopper is the Sketch feature—you can doodle something rough, add a text prompt, and watch it transform into a polished image. There's also a nifty annotation tool for precise edits, plus two API options for developers. Curious about the details? Read on to see what's new and how it might change your creative workflow.

DeepSeek has announced a price cut for its Flash series models, effective September 10, 2026. The new pricing reduces costs across the board, with output dropping to just 4 yuan per million tokens during off-peak hours. Cache hits see the steepest reduction at 60%, making AI more affordable for developers and businesses.

OpenAI has announced that its internal AI model solved the Navier-Stokes existence and smoothness problem, one of the Clay Mathematics Institute's Millennium Prize Problems. The breakthrough, achieved by 10,000 AI agents running for 88 hours, has excited the math community but also raised concerns about the role of human understanding and the ethics of using others' research. Mathematician Terence Tao praised the achievement while warning about its implications.

The first AIGC-adapted wuxia film, 'Shan Zhu Dao,' has launched on Tencent Video and Jiguang TV, blending AI technology with traditional Eastern aesthetics. Based on a popular comic, it tells the story of a female swordsman's journey of redemption. The production overcame common AI pitfalls to deliver a visually stunning and emotionally resonant experience.

近日，多位Claude用户发现自己的账户额度在未使用的情况下被迅速消耗。Anthropic调查后确认，黑客通过恶意软件窃取登录会话，非法使用用户的付费额度。英国用户Grant D'Arcy成为首个受害者，他的账户在一天内被消耗了10%的额度，最终获得退款。事件在Reddit和GitHub上引发热议，更多受害者浮出水面。Anthropic提醒用户检查设备是否感染恶意软件，但部分用户未收到安全提醒。

Runway, the AI startup known for its video generation tools, has crossed $200 million in annual recurring revenue. But the company is pivoting away from the hype of text-to-video towards 'world models'—AI systems that simulate physical environments. With major clients like Amazon and Robinhood, Runway is betting on long-term enterprise contracts over fleeting trends. This strategic shift could redefine its future in the competitive AI landscape.

讯飞正式推出星火X2.5大模型，采用混合专家架构，总参数达2930亿，但推理时仅激活300亿，兼顾性能与效率。新模型在代码生成、智能体交互、数学计算和语义理解等方面均有显著提升，并支持256K长上下文，针对国产算力环境深度优化。同时，讯飞还开源了4B和1.7B两款轻量级端侧模型，支持百万级上下文，适合本地部署。目前，星火X2.5系列已在讯飞开放平台上线，并推出限时优惠，为开发者和企业提供更具性价比的选择。