Discover the AI Tools

ByteDance pulled in 30% more revenue in the first half of 2026, but its profit margin slipped to 16.7%. Why? The company is pouring money into AI — and it's not alone. Overseas revenue hit a record high, while a $30 billion bank loan shows just how high the stakes have become.

Meta 推出 Meta One 订阅服务，个人版每月 7.99 美元起，企业版最高 499 美元。服务整合 Facebook、Instagram、WhatsApp 的 AI 功能与商业工具，标志着 Meta 在 AI 商业化上迈出重要一步。

OpenAI is eyeing a $1.2 trillion valuation in new funding talks, but it's in no rush to go public. CEO Sam Altman says $1 trillion is the floor for an IPO, now pushed to 2027. With soaring revenue, widening losses, and fierce competition, the real question is whether this sky-high bet will pay off.

OpenAI has announced that GPT-5.5 will be discontinued on October 14, affecting ChatGPT, ChatGPT Work, and Codex across all subscription plans. Codex users relying on GPT-5.5 are urged to switch to GPT-5.6 Sol or GPT-6 Astra. The model, launched in April, was praised for reducing token consumption, but its retirement signals a rapid shift in AI offerings.

At the All-In summit, Elon Musk proposed a bold idea: top AI companies should cross-test each other's models before release. He argues that models smart enough to be useful are also smart enough to break free—and only competitors can catch the flaws. Drawing on the Hugging Face incident, Musk says even Anthropic's own staff fear their creations. Could peer review be the safety net we need?

Anthropic just dropped a beta that plugs Claude right into Salesforce. Imagine having a super-smart assistant that knows your customers, deals, and pipeline inside out. It can even update records—with your okay. Early users at GitLab, Siemens, and Legora are already giving it a spin. Could this be the productivity boost sales folks have been waiting for?

Google's new Gemini 3.8 Live and Live Extended Thinking models can reason, search, and run tasks in the background—all while you're still speaking. They auto-detect 97 languages, switch mid-conversation, and even tell you 'Let me check...' when they need a moment. Early benchmarks look strong, and a SynthID watermark keeps AI audio honest.

Google just unveiled Gemini 3.8 Live, a voice model that finally cracks the code on real-time conversation and deep reasoning. No more awkward silences while the AI thinks—this one keeps the chat flowing while it works through complex problems in the background. Developers can start building with it today.

On September 15, the Chinese Internet Basic Corpus 4.0 quietly went live in Jinan, packing 120 gigabytes of high-quality data to help train better large language models. Built by the China Cybersecurity Association and partners like Baidu and iFLYTEK, it's now open for download. Here's what it means for the future of Chinese AI.

Ant Afu just rolled out an AI diet score that analyzes your meal from a single photo. It also added 100 million eggs to its 'Scientific Weight Loss of 100 Million Jins' campaign. Users who snap three meals for three days can grab eggs for a penny. Since June, over 2.75 million users have lost more than 5 million jins combined, with AI trackers doubling average weight loss. Sounds egg-cellent, right?

Salesforce just dropped its first reasoning model, Koa, built with NVIDIA's Nemotron. It's open-source, trained without real customer data, and promises to slash costs for businesses. Can it really challenge the closed-source AI titans?

A leaked document reveals OpenAI's internal 'Project Lily,' where human reviewers read anonymized user chats to improve ChatGPT. While the goal is to make AI sound less robotic, the practice raises privacy red flags. Users' intimate conversations might not be as private as they think—and OpenAI's response has been less than transparent.

Ever handed your phone to someone who just can't frame a shot? Melody Chu felt that pain. The former TikTok exec built Superpose, an AI camera that suggests poses instead of swapping backgrounds. In two months, it's been downloaded over 22,000 times, with users snapping nearly 190,000 poses. The startup just raised $2.2 million in seed funding.

At its 2026 developer conference, vivo unveiled four upgraded Blue Heart AI models. They can understand speech in noisy places, remember your habits right on your phone, and even control your device to get things done. It's like giving your phone a brain that actually pays attention to you.

China's industrial AI model sector is booming. New data shows over 62,000 companies, with a whopping 22,300 added this year alone. Guangdong, Jiangsu, Zhejiang, Shanghai, and Beijing lead the charge, thanks to their strong industrial and tech foundations. This surge signals deep integration of AI with the real economy.

HP just unveiled the ZBook Ultra G3a 16, a mobile workstation that packs a jaw-dropping 192GB of unified memory. What does that mean for you? It can run AI models with up to 300 billion parameters locally, no cloud required. At $7,449, it's a serious tool for pros who need raw power and data privacy.

Huawei's new AI assistant, Xiao Yi Work, is now in internal testing. It lets you run tasks on multiple computers remotely, even when you're away. The service handles everything from coding to creative design, and it's available on HarmonyOS devices plus Windows. But there's a catch: free user data vanishes after expiry. Curious how it works? Read on.

Apple's new Reference Image mode on iPhone 18 Pro and Pro Max signs photos at the moment you press the shutter, using the Secure Enclave to create a tamper-proof 'digital negative.' It's a direct counter to AI-generated fakes, aimed at journalists and anyone needing rock-solid image proof. But will it change how we trust what we see?

In a candid internal memo, Microsoft CEO Satya Nadella argues that AI must stay under human control. He calls for broader third-party testing and warns that companies must take as much time as needed to ensure safety—or risk losing their operating licenses. The message comes as industry leaders debate slowing down AI development.

Shanghai AI Lab just dropped Shusen-S2, an open-source multimodal model that's turning heads in the AI community. With a plug-and-play memory module and top-tier performance in scientific tasks like biology and chemistry, it's challenging closed-source heavyweights. Could this be the open-source breakthrough researchers have been waiting for?