Discover the AI Tools

Anthropic has unveiled the Model Hardware Standard (MHS), a new protocol that lets AI control physical devices like robotic arms and lab equipment. Early tests show it can slash experiment times by two-thirds, marking a bold step from chatbots to hands-on automation. This move signals a shift in AI competition from conversation to real-world action.

Anthropic has unveiled MHS, a new standard that enables AI agents to safely operate physical devices. Developed with HHMI's Janelia Research Campus, MHS cuts hardware integration time from weeks to minutes. Early tests show Claude exhibiting exploratory behavior akin to human scientists, autonomously reasoning and recovering from failures. While limitations exist, MHS marks a significant step toward bridging AI's virtual and physical worlds.

腾讯混元于8月28日正式发布开源旗舰模型Hy4preview，总参数达7700亿，激活参数490亿，支持100万上下文长度。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，并已在HuggingFace、GitHub等平台开源。内部盲测显示，其工程任务评分略优于GLM-5.3和Kimi K3。腾讯表示，这是Hy4的早期版本，将根据反馈快速迭代，标志着大模型竞争从参数比拼转向实际生产力验证。

Ant Group, in collaboration with CICC, has launched Ling-3.0-flash-Fin, the first financial-enhanced large model under Ant Bailing. This model significantly boosts research and investment capabilities through specialized training on financial data. It excels in information retrieval and general tasks, and will be open-sourced soon. Developers can access free API calls for a month on OpenRouter, with model weights to follow next week.

Weibo has permanently closed accounts that spread fake chat records between Sun Yuchong and Eileen Gu, which were generated by AI. The rumors, exposed by Sun's ex-girlfriend Zeng Ying, were part of a suspected black-industry promotion scheme. Weibo's action highlights the growing challenge of AI-generated misinformation on social media.

腾讯混元今日发布新一代旗舰模型Hy4preview，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，多项盲测超越同类竞品。目前该模型已全面开源，并上线腾讯云TokenHub和OpenRouter，用户可通过元宝、ima等产品体验。

腾讯今日发布混元Hy4预览版，拥有770B总参数和49B激活参数，上下文长度达1M。该模型在软件工程、办公分析、游戏开发等场景表现亮眼，盲测得分超越GLM-5.3和Kimi K3。同时，在三维Blaschke-Lebesgue几何问题上取得突破。模型已开源，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最新开源旗舰Hy4preview，国内外版本同步上线，并推出限时免费策略：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、49B激活参数，上下文长度达1M，主打编程、办公和科学等生产力场景。开发者现可在WorkBuddy中直接调用，实现从模型发布到产品落地的快速集成。此举旨在降低门槛吸引开发者，同时延续混元通过开源和生态协作推动大模型普及的思路。

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 the tech industry's approach to AI development. This investment highlights the growing reliance of tech giants on third-party AI expertise and the intensifying competition in the large model arena. What does this mean for the future of AI innovation?

Google has just dropped a new AI video model, Gemini Omni 1.1 Flash, that can generate stunning 4K videos. It's not just about resolution—this tool lets you extend clips seamlessly, control camera movements, and even preview in 360p for lightning-fast drafts. Whether you're a developer or a creator, this update is a game-changer. Let's dive into what makes it tick.

Google has unveiled Gemini Omni 1.1 Flash, a video generation model that can produce clips up to 4K resolution. It offers a full pipeline from quick storyboard previews to final renders, with features like scene expansion and reference video support. While higher resolutions cost more and take longer, the model aims to make AI video production more practical for professionals.

OpenAI, Microsoft, Google, and 113 other companies have signed a joint letter calling for urgent action on cybersecurity. They warn that AI-powered attacks will surge in the coming months and urge governments to fast-track trusted access programs and fund critical infrastructure defenses. The letter also highlights a recent breach at Hugging Face as a wake-up call.

NVIDIA has quietly shelved a financing scheme for AI cloud providers that would have traded credit for revenue shares. Insiders say the move aims to sidestep antitrust scrutiny. The plan, which also included buyback guarantees, raised red flags internally. Will this strategic retreat help NVIDIA navigate regulatory minefields?

Baidu's switch to a dual primary listing on the Hong Kong Stock Exchange takes effect September 1, 2026, making it the first company with a full-stack AI layout listed on both Nasdaq and HKEX. This move, which doesn't involve new share issuance, is expected to attract southbound funds and boost liquidity. With AI revenue now half of its total, Baidu's valuation could shift from traditional metrics to a sum-of-the-parts approach, potentially unlocking significant value.

OpenAI联合苹果前高管及Jony Ive的公司，向法院申请永久驳回苹果的诉讼，指控苹果的起诉缺乏具体事实。这场法律战不仅关乎人才流动，更可能影响硅谷的技术竞争格局。

Google's AI Mode now helps you track flight prices and book hotels across 300+ platforms. With real-time data from airlines and travel sites, you can set price alerts and even use miles or points. This update turns AI Mode into a full-fledged travel agent, making trip planning easier than ever.

In a bold move that blurs the lines between quantum physics and artificial intelligence, China's Guoguang Quantum AI team has unveiled 'Xuanmi', the country's first large language model enhanced by quantum technology. Designed for scientific and academic use, it promises sharper reasoning and faster decisions in complex tasks. But what does this mean for the future of AI? Let's dive in.

小鹏汽车近日发布了二代VLA大模型的首次重大升级，新系统XOS 6.3.0让AI真正理解时间，从静态3D空间理解跃升到动态4D时空理解。通过增大模型参数、引入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，车机端首次引入Master Agent，实现VLA与VLM融合，并推动L4级Robotaxi技术下放量产车。

Midjourney 正式推出 V8.2 图像编辑模型，打破传统文生图单一模式，支持指令编辑、多图参考、局部重绘和画布扩展。新模型还兼容个性化设置与风格参考，为创作者提供更精准可控的工具。目前处于测试阶段，官方邀请用户反馈问题。