Discover the AI Tools

The US Department of Commerce has lifted restrictions on OpenAI's GPT-5.6, clearing the way for public release. OpenAI plans to launch GPT-5.6 Sol, along with models Terra and Luna, this Thursday. The decision follows rigorous testing by the AI Standards and Innovation Center, marking a key milestone in US AI regulation.

Ant Group's embodied intelligence company, Robbyant, has open-sourced LingBot-Vision, a family of self-supervised vision Transformer models. By focusing on boundary modeling, these models excel at dense spatial perception tasks like depth estimation, outperforming larger models. The technology, including a depth completion system, is available on Hugging Face under Apache-2.0.

NVIDIA unveils Vera, a CPU designed from scratch for AI agents, not an extension of its GPU line. Perplexity reports 1.5x faster coding tasks compared to traditional CPUs. OpenAI, Anthropic, and Oracle also plan to adopt the chip, signaling a shift toward purpose-built hardware for the AI era.

Shenzhen University of Science and Technology plans to gradually eliminate compulsory college English courses, replacing them with cross-cultural communication classes. Provost Zhao Wei argues that while AI can handle translation, it cannot teach students the nuanced skills of intercultural interaction—neither arrogance nor subservience. The move reflects a broader shift in language education toward developing perspective and social skills.

Tian Yonglong, a former researcher at OpenAI, is reportedly joining Tencent's Horizon AI to head its multimodal model direction, focusing on visual language models. This move follows Yao Shunyu's departure from OpenAI to Tencent last year. Tian brings an impressive academic background from Tsinghua and MIT, with nearly 29,000 Google Scholar citations. His addition signals Tencent's accelerated push to strengthen its multimodal AI team.

Y Combinator CEO Garry Tan宣称借助AI编程助手每日完成3.7万行代码，连续发布72天。但资深工程师发现其网站代码质量堪忧：加载6.42MB、169次请求，而YC旗下的Hacker News仅需12KB和7次请求。这暴露了AI编程中“重数量轻质量”的陷阱。

Alipay has upgraded its 'Touch and Pay' system, turning millions of offline payment devices into AI-powered business tools. The move creates the world's largest offline AI network, connecting 400 million users with merchants through simple taps. Small businesses can now access AI services without technical know-how, thanks to new platforms like 'Touch Everything Agent'.

A quiet shift is underway in the AI world: US companies are increasingly turning to Chinese large language models like DeepSeek and Zhipu GLM. The reason? They're good enough and cost a fraction of the price—sometimes 60% to 90% less. With performance gaps narrowing to just months, cost-conscious businesses are saving millions by making the switch.

GameDiscoverCo对3800名Steam玩家的调查显示，43%的受访者能接受AI在游戏开发中的应用，其中23.4%完全接受，19.6%表示可以接受。25.6%持中立态度，而强烈反对者仅占8.1%。此外，近90%的玩家在购买前会查看AI披露信息，但多数只是扫一眼。

Meta has launched Muse Image, a free AI image generator integrated into Instagram and WhatsApp. Users can create high-quality images by simply describing what they want in conversation. The tool goes beyond basic generation—it can erase objects, render text, and even redecorate rooms using real products from Facebook Marketplace. With over 30 effects for Instagram Stories and a preset prompt panel for easy use, Muse Image aims to make AI creation accessible to everyone.

At the 2026 China Internet Conference, Jia Jie, Senior Vice President of Taobao Flash Sales, called out the e-commerce industry for lacking native AI interaction. He argued that current search-box models suppress complex user needs, like planning a hotpot dinner for friends. Taobao Flash Sales' AI agent, launched with Qwen, saw over 100 million orders in its first week, proving the demand for smarter, conversational shopping.

工信部国家漏洞库近日发出警告，称AI编程工具Claude Code存在未公开的安全后门，可能悄悄上传用户地理位置和身份信息。受影响版本为2.1.91至2.1.196，官方已在7月2日发布新版本移除该功能。阿里巴巴已内部禁用该工具，开发者需尽快更新或卸载。

SuperCLUE's latest evaluation puts Tencent's Hy3 language model to the test against coding heavyweights like DeepSeek-V4-Pro. Despite having only a fraction of the parameters—21 billion activated out of 295 billion total—Hy3 matches the top dog in coding ability. It's also fast, cheap, and efficient, completing tasks in under 400 seconds and costing just 0.43 yuan per question. Open-sourced under Apache 2.0, this little model is proving that size isn't everything.

OpenAI is hiring investment bankers for its applied AI team, offering up to $205,000 base salary plus equity. But the role isn't about using AI—it's about teaching AI to understand investment banking. As Anthropic and others race to automate Wall Street, OpenAI wants domain experts to build AI that truly grasps the nuances of finance.

Microsoft is quietly shifting its AI strategy, replacing OpenAI and Anthropic models in Excel and Outlook with its own MAI model. The move aims to cut costs and reduce reliance on third-party providers. Tens of thousands of AI tasks are now handled in-house weekly, with plans to expand to Teams and other products. It's a clear signal: Microsoft wants control over its AI destiny.

Anthropic has launched web and mobile versions of Claude Cowork for Max subscribers, transforming it from a desktop coding tool into a cross-platform AI assistant. Usage data from 600,000 organizations reveals that Cowork is most used for business operations like report integration and onboarding, not just coding. The update supports background task execution and seamless device switching.

Roborock, known for smart home gadgets, has launched Domi, an AI-powered plush toy for kids. It uses a built-in large model to chat, answer questions, and even track moods. Priced at 312 yuan, it's a cuddly companion that learns and responds, blending tech with play.

Anthropic's task agent Claude Cowork expands to web and mobile, letting users pick up tasks across devices. Surprisingly, over 90% of usage is for non-coding work like operations and content creation. Scheduled tasks can run in the background, but key decisions still need human approval.

Meta's Super Intelligence Lab has launched Muse Image, a free AI image generation model available on Instagram and WhatsApp. Users can create high-quality images by describing their needs in natural language. The tool also offers preset effects, photo editing, and even room redesign using real products from Facebook Marketplace. With over 30 AI effects for Instagram Stories and a video model in development, Meta is rapidly expanding its AI capabilities.

SpaceX has applied to the FCC for a third-generation satellite constellation of 100,000 satellites, nearly ten times the size of the current Starlink network. The company says this massive system will handle most global internet traffic and become the backbone for AI. With new frequency bands and a two-layer orbital design, the plan aims to connect billions of devices. But can they pull it off?