Discover the AI Tools

Anthropic has unveiled the Model Hardware Standard (MHS), a new protocol that lets AI agents control physical devices like robotic arms and lab equipment. This move marks the company's entry into embodied AI, aiming to bridge the gap between digital intelligence and the physical world. Early tests show dramatic efficiency gains, such as cutting experiment time by two-thirds.

Anthropic has unveiled MHS, a new standard that enables AI agents to operate physical devices safely and efficiently. Developed with HHMI's Janelia Research Campus, MHS cuts hardware integration time from weeks to minutes, allowing AI to autonomously run experiments and recover from failures. While still in preview, it marks a significant step toward bridging the digital and physical worlds.

腾讯混元于8月28日开源了旗舰模型Hy4preview，总参数770B，激活参数49B，支持1M上下文。该模型在软件工程、办公、游戏开发和科学研究四大领域均有突破，尤其在三维Blaschke–Lebesgue问题中取得了重要进展。腾讯表示，这是Hy4的早期版本，将通过快速迭代吸收反馈。

蚂蚁集团与中金公司合作，正式发布首个金融增强大模型Ling-3.0-flash-Fin。该模型在原有基础上，通过专业金融数据训练，大幅提升了投研能力，并在信息检索等四大领域表现卓越。同时，双方将开源金融搜索基准FinFIRST，并计划下周开放模型权重，为开发者与金融机构提供便利。

Weibo has permanently closed accounts that used AI to fabricate chat records between Sun Yuchong and Eileen Gu. The platform confirmed the content was entirely fake, and the rumor-mongering accounts were part of a suspected black-market traffic scheme. This incident highlights the growing challenge of AI-generated misinformation and the need for stricter platform oversight.

腾讯混元今日发布新一代旗舰模型Hy4preview预览版，总参数量达770B，激活参数49B，支持1M上下文。在编程、办公、科研等真实生产力任务中表现出色，多项内部盲测超越同类竞品。该模型现已全面开源，并上线腾讯云TokenHub和OpenRouter，用户可通过元宝、ima等产品体验。

腾讯今日开源了混元Hy4预览版，拥有770B总参数和49B激活参数，支持百万级上下文。官方称其在规模、上下文和数据量上均有显著提升，智力水平再上台阶。在内部盲测中，Hy4预览版得分略超GLM-5.3和Kimi K3，并在三维几何难题上取得突破。目前该模型已同步上线HuggingFace、GitHub等平台，并集成至腾讯云TokenHub和OpenRouter。

腾讯元宝于8月28日宣布接入混元大模型Hy4preview，国内与海外版本同步上线，并推出限时免费政策：Hy4preview提供两周免费试用，前代Hy3免费期延长至9月30日。Hy4preview拥有770B总参数和49B激活参数，支持1M上下文，定位编码、办公、科研等生产力场景。此举旨在吸引开发者零门槛体验，收集反馈，同时延续混元通过开源和生态合作推动大模型普及的策略。

Hugging Face has unveiled Microduck, a 25-centimeter open-source duck robot that can waddle, carry up to 800 grams, and even roller skate. Priced at $399 and shipping before Christmas, it's designed to make embodied AI accessible to developers. With cameras, lidar, and IMUs, it supports reinforcement learning in simulation. The move underscores Hugging Face's push into open-source hardware, amid reports of a potential $13 billion acquisition by NVIDIA.

Meta is reportedly planning to spend up to $10 billion annually on Anthropic's AI models, signaling a major shift in how tech giants approach AI development. This investment highlights the growing reliance on top-tier third-party models and the intensifying competition in the AI landscape. What does this mean for the future of AI innovation?

Google just dropped Gemini Omni 1.1 Flash, a video generation model that cranks out 4K footage. It's not just about resolution—you can extend clips, control camera moves, and even use reference videos. Plus, a nifty 360p mode speeds things up by 60% for quick drafts. Here's what creators need to know.

Google has unveiled Gemini Omni 1.1 Flash, an AI video model that cranks resolution up to 4K. It's a game-changer for creators, offering scene expansion, precise frame control, and a tiered pricing structure that makes high-end video production more accessible. From quick storyboards to polished final cuts, this tool bridges the gap between concept and delivery.

OpenAI, Microsoft, Google, and over a hundred other companies have signed a joint letter warning that AI-powered cyberattacks are about to surge. They're calling on governments and businesses to bolster defenses now, before it's too late. The letter proposes a 'Trusted Access Program' and emphasizes protecting critical infrastructure like hospitals. It's a stark reminder that as AI gets smarter, so do the threats.

NVIDIA has quietly shelved a financing program for AI cloud providers that would have taken a cut of their revenue. The move comes amid internal concerns about antitrust scrutiny and the company's growing influence over customers. The plan, set to launch in July, was halted just weeks before its debut, raising questions about NVIDIA's strategy in the AI computing market.

Baidu's switch to a dual primary listing on the Hong Kong Stock Exchange takes effect September 1, 2026, making it the first Nasdaq and HKEX listed company with a full-stack AI layout. This move, without new share issuance, is expected to attract southbound funds and boost liquidity. With AI revenue now half of its total, Baidu's valuation may shift from traditional metrics to a sum-of-the-parts approach, potentially unlocking significant value.

OpenAI, along with former Apple executives and Jony Ive's design firm, has filed a motion to permanently dismiss Apple's trade secret theft lawsuit. The legal battle highlights the fierce competition for top tech talent and raises questions about intellectual property in the AI era. A court hearing is set for October 1st.

Google has supercharged its AI Mode with real-time flight tracking and hotel booking across 300+ platforms. Now you can ask for travel plans in plain language, get price alerts, and even redeem miles or points—all without leaving the search bar. This move turns AI Mode into a full-fledged travel agent, reshaping how we plan trips.

China's Guoguang Quantum AI team has unveiled Xuanmi, the first quantum-enhanced large language model tailored for science and academia. By weaving quantum tech into decision-making and reasoning, Xuanmi promises sharper logic and faster paths for complex tasks. Could this be the bridge to a new era of computing?

小鹏汽车近日发布了第二代VLA大模型的首次重大升级，新版本XOS 6.3.0在G9L上全球首发。这次升级让AI真正理解时间，从静态3D空间理解跃升至动态4D时空理解。通过增加3.5倍参数、引入超长时序记忆和流式推理，端到端响应速度提升300%，综合安全能力提升20倍。同时，车内首次引入Master Agent，实现VLA与VLM融合，并带来语音控车等新功能。

Midjourney 正式推出 V8.2 图像编辑模型，打破纯文本生成限制，支持指令编辑、多图参考、局部重绘和画布扩展。用户可通过拖拽图片或点击编辑按钮轻松使用，官方已开放测试并邀请社区反馈。