Discover the AI Tools

As college entrance exam results roll in, Qwen has partnered with China Post EMS to offer real-time tracking and proactive notifications for admission letters. Users can authorize the service to receive updates when letters are dispatched and view logistics on the interface. The feature builds on Qwen's existing volunteer calendar, which has already helped millions of students navigate the application process.

London-based AI film company Particle6 has begun production on 'Misaligned,' a feature film starring their AI creation Tilly Norwood. The film explores an AI's journey to humanity, but Hollywood isn't happy. Unions and stars like Emily Blunt criticize AI actors as a threat to human artistry. Is this the future of cinema or a step too far?

Meta is developing smart glasses that can record audio and snap photos every few seconds, all day long. The device uses a 'super perception' mode to capture your surroundings, letting you ask questions about what it sees and hears. But privacy concerns are already boiling over, especially since the recording light might stay off during active sensing.

Step Star, a Chinese AI unicorn, is launching its first AI smartphone, beating OpenAI to market. The phone, manufactured by Huaqin Technology, represents a shift from pure cloud AI to integrated hardware-software agents. With over 20 billion yuan raised and a Hong Kong IPO in sight, Step Star is betting big on native AI terminals.

French AI company Mistral has released Robostral Navigate, an 8B-parameter model that enables robots to navigate autonomously using only a single RGB camera. In tests, it outperformed multi-camera and depth-sensor systems, achieving over 76% success in new environments. The model was trained entirely in simulation and works with wheeled, legged, and flying robots.

Amazon is quietly developing an AI agent called 'Moonraker' that will let Alexa handle complex, multi-step tasks—like booking a ride and texting a friend—all from a single command. The project is costly, with GPU expenses topping $100 million in 2026, but early signs show Alexa Plus usage has doubled and orders tripled. It's a high-stakes bet to turn Alexa from a simple voice assistant into a true digital helper.

毕马威最新调查显示，近三成企业高管坦言搞不懂AI的经济账，按量计费模式让成本变得像黑箱。当AI不再像月订阅那样简单，企业正从“买工具”转向“算细账”，而那些真正理解AI成本逻辑的公司，将在下一轮竞争中抢占先机。

OpenAI has publicly challenged SWE-Bench Pro, a popular benchmark for evaluating AI coding skills, claiming that about 30% of its 731 test tasks contain flaws. The company points to an unusually rapid rise in pass rates—from 23% to 80% in just eight months—as evidence that the benchmark no longer measures true ability. OpenAI has withdrawn its recommendation of the benchmark and calls for a new evaluation system designed by experienced developers.

OpenAI has launched GPT-Live, a full-duplex voice mode for ChatGPT that lets you interrupt, pause, and get real-time feedback—just like talking to a human. No more waiting for the AI to finish before you speak. The update uses the GPT-5.5 model and is rolling out to paid and free users.

Lovable, a Swedish startup behind 'vibe-coding' tech, is in talks to raise $3 billion, potentially doubling its valuation to $13.2 billion. The company, founded less than three years ago, hit $500 million in annualized revenue in June. Its AI tool lets users build software by describing it in plain language, attracting everyone from indie designers to giants like Nvidia.

Google Photos just launched 'Video Remix,' a new AI-powered editing tool that lets anyone turn ordinary videos into cinematic masterpieces in seconds. Powered by the Gemini Omni processor, it can brighten dark scenes, swap backgrounds, or apply artistic filters like watercolor and oil painting. Rolling out to subscribers in over a dozen countries, this move pits Google against Apple, OpenAI, and Adobe in the race to bring generative AI to everyday users.

Tencent's new Hy3 AI model, launched on WorkBuddy, was so popular that it overwhelmed computing resources, causing queues and delays. The team scrambled to expand capacity, and now it's back on track. The model boasts a 90% task success rate and 34% faster task completion, drawing huge demand from office users and developers.

Elon Musk's SpaceX AI just dropped Grok 4.5, a new model that Musk claims rivals Anthropic's top-tier Opus in capability but costs a fraction. With double the token efficiency and aggressive pricing—$2 per million input tokens—Grok 4.5 directly challenges OpenAI and Anthropic. The model goes public tomorrow after strong trial feedback, heating up the AI arms race.

Microsoft has invited Insider members to test a new iPad version of Word that deeply integrates the Copilot AI assistant. The update aims to streamline document editing on mobile devices, allowing users to transform scattered ideas into structured documents and optimize text tone and layout. However, the current test version has limitations, such as only modifying existing files and lacking image generation and comment editing features.

SpaceX AI, in collaboration with coding startup Cursor, has released Grok 4.5, a new AI model designed for complex, long-running tasks in law and finance. The model aims to close the gap with top-tier systems from OpenAI and Anthropic, offering enhanced cybersecurity and versatility beyond software engineering.

Step Star, a unicorn in the large model space, is set to release the world's first AI agent phone from a major AI model vendor. The device, powered by a new intelligent agent system, aims to transform smartphones from passive tools into proactive assistants. This move positions Step Star ahead of rivals like OpenAI, which plans a similar launch in 2027, and could reshape the mobile industry.

AI startup Anthropic has confidentially filed for an IPO, potentially becoming the largest publicly traded AI lab. With profits expected to exceed $1 billion this quarter and a combined ARR with OpenAI nearing $100 billion, Anthropic is racing ahead while OpenAI delays its own IPO until 2027. The move could reshape the competitive landscape, forcing rivals to disclose finances and accelerate fundraising.

ByteDance just dropped Seedream 5.0 Pro, a multimodal image model that goes beyond simple generation to offer interactive precision editing. It can turn complex data into sleek infographics, let you tweak images by circling objects, and even handle multilingual text. Think of it as a design assistant that actually gets what you want—no more endless back-and-forth.

Cloudflare and OpenAI have launched a pilot project that uses Cloudflare's network data to make AI search engines faster and more accurate. By tapping into real-time signals like content update frequency and traffic trends, AI models can now better understand what's happening on the web right now. This could mean fewer outdated answers and more reliable responses when you ask AI about current events.

Yonglong Tian, a former researcher at OpenAI, has joined Tencent's large language model department to focus on visual language model development. The move highlights Tencent's push to strengthen its multimodal AI capabilities amid fierce industry competition.