Discover the AI Tools

Insta360 is rolling out an AI voice assistant for its GO Ultra thumb camera, integrating Alibaba's Qwen in mainland China and Google's Gemini elsewhere. The assistant, activated by saying "Hey Kira" or long-pressing the capture button, supports face-to-face translation, photo-based Q&A, and voice commands. Founder Liu Jingkang hints this could redefine thumb cameras, signaling AI's push into consumer hardware.

Google's Creative Lab has unveiled Gemma Translator, a device that runs a 5.1-billion-parameter AI model on a Raspberry Pi 5, enabling real-time, offline translation. The project showcases how edge AI can move beyond labs into practical, standalone products. With speech recognition and synthesis handled locally, users can converse across languages without any internet connection, making it a promising tool for travelers and remote areas.

Stanford and Arc Institute researchers used the AI model Evo to generate 700,000 viral genomes, synthesized 285, and found 16 that could infect and kill E. coli. This marks a leap in generative biology but raises biosecurity concerns. The study, published in Science, shows AI can now design entire functional viruses, not just proteins. While the viruses are harmless to humans, experts worry about potential misuse.

Insta360's GO Ultra thumb camera now has an AI voice assistant, using Alibaba's Qwen in mainland China and Google's Gemini elsewhere. It supports face-to-face translation, image Q&A, and voice Q&A, turning the tiny camera into a personal AI companion. Founder Liu Jingkang believes this could redefine the thumb camera and potentially become a top-selling personal AI device.

Neon and Castform have trained a 4B-parameter model that outperforms GPT-5.6 in document search accuracy while costing just 1% per inference. This breakthrough challenges the notion that bigger is always better, and it's reshaping how we think about AI search. By using reinforcement learning and a clever division of labor, they've made agent-based search both smarter and more affordable.

Xiaomi's latest smart camera isn't just about watching—it understands. The 4 Max AI Zoom Edition, priced at 799 yuan, packs an AI model that can tell if your kid is studying, if grandma has fallen, or if your cat hasn't eaten. But the AI features come with a subscription after a three-month trial. Is it worth it? We break down the specs and the fine print.

Volcano Engine has rolled out the Seedance 2.5 API, bringing significant upgrades in instruction following, long-form storytelling, and visual realism. The new model supports native 30-second video output, up to 50 multimodal references, and enhanced editing precision. With discounted pricing for older versions, this move aims to make AI video creation more accessible and commercially viable.

At a recent roadshow, Unitree's founder and CEO, Wang Xingxing, shared his vision for the future of robotics. He emphasized the company's commitment to embodied intelligence, with plans to expand beyond quadruped robots into humanoid robots and mechs. The goal? To let robots handle the heavy lifting and risky jobs, freeing up humans for more creative pursuits. As the industry heats up, Unitree is positioning itself as a full-stack player, from core components to AI models.

The AI arms race is getting expensive, and tech giants are turning to debt to fund their ambitions. Alphabet is planning to raise up to $25 billion through bond sales, while SoftBank is borrowing $10 billion using its OpenAI shares as collateral. These moves highlight the massive capital needed for AI infrastructure and the high-stakes bet that these investments will pay off.

ByteDance is reportedly in early talks to train a massive AI model with over 5 trillion parameters, which would surpass Alibaba's and Moonshot AI's largest models. The project, led by Seed Foundation's Xiang Liang, is still unconfirmed but signals a shift toward foundational model competition in China.

OpenAI 宣布取消 ChatGPT 的文本聊天限制，所有用户均可无限制使用。新推出的 GPT-5.6 系列模型在事实错误率上大幅降低，并引入了“思考”按钮和“思考级别”滑块，让用户能更灵活地控制 AI 的推理深度。这一举措标志着 ChatGPT 正从简单的问答工具，向高频、智能的交互平台迈进。

OpenAI is reportedly developing its first AI hardware device—a doughnut-shaped smart speaker without a screen, priced between $300 and $400. Designed with premium materials and unique moving parts, it aims to bring ChatGPT into everyday spaces. But with the smart speaker market's profit challenges and a looming Apple lawsuit, will this device succeed? Here's what we know so far.

OpenAI is set to release its new flagship model, Astra, next week. Leaks suggest it's the largest pre-training since GPT-4.5, with a focus on multi-agent collaboration. Astra has reportedly solved 10 open math problems, hinting at its advanced reasoning. This launch comes amid fierce competition from Google and Anthropic, positioning Astra as a potential game-changer for complex tasks and enterprise AI.

OpenAI's debut hardware is shaping up to be a screenless smart speaker, roughly the size of a hockey puck, priced between $300 and $400. Designed with Jony Ive's LoveFrom, it packs a camera, battery, and a self-moving component, aiming to be an always-on AI companion. But it arrives amid Apple's trade secret lawsuit, adding a layer of intrigue.

AI music platform Suno is taking steps to curb misuse of its generated tracks. The company will add watermarks, restrict downloads, and update community guidelines to prevent counterfeit songs. These moves come amid lawsuits from record labels and a recent court ruling against Suno in Germany. Suno also partnered with Musixmatch for copyright detection, aiming to balance creativity with responsibility.

Yushu Technology's IPO strategic placement revealed DeepSeek's 933,399-share allocation, worth about 141 million yuan, with a 36-month lock-up. Tencent also participated. The company's valuation hits 60.9 billion yuan, but high valuation ratios raise concerns. This marks a milestone for embodied AI investments.

OpenAI has rolled out a major ChatGPT update, upgrading free users to the GPT-5.6 Luna model and lifting chat limits, while paid subscribers get a new slider to control thinking depth. This move aims to attract more users and fend off rivals like Google Gemini. The update also introduces a 'Think' button for on-demand advanced reasoning, giving everyone more flexibility. Here's what changed and why it matters.

OpenAI has rolled out a major update to ChatGPT, giving free users unlimited text chats powered by the new GPT-5.6 Luna model. Plus and Pro users aren't left out—they get an optimized GPT-5.6 Sol with better accuracy and a new slider to control thinking depth. The update also cuts factual errors significantly, making ChatGPT smarter and more reliable for everyday questions and complex tasks alike.

字节跳动正计划训练一个参数超过5万亿的大模型，规模远超阿里Qwen 3.8-Max和月之暗面K3，有望成为国内之最。创始人张一鸣罕见发声，强调长期主义，并下令内部禁止蒸馏开源模型。CEO梁汝波在全员会上重申自研决心，称这是为了夯实技术基础，而非外部压力。这一系列动作，既是技术路线之争，更是战略宣言。

最新披露的数据显示，OpenAI已成为微软AI业务的重要支柱。在截至今年6月的财年中，OpenAI为微软贡献了约241亿美元收入，占其AI总收入的70%。这一合作既带来了可观收益，也让微软在战略上开始寻求平衡，逐步减少对外部模型的依赖，加速自研AI模型在核心产品中的整合。微软与OpenAI的关系正从单纯合作走向竞合并存。