Group 1 - Tesla has announced that the design of its AI5 chip is nearing completion, with the AI6 chip in early stages, aiming to shorten the chip design cycle to 9 months and predicting it will become the highest production AI chip globally [1] - The AI5 chip will utilize Samsung's 2nm and TSMC's 3nm processes, boasting overall performance 50 times that of AI4 and memory capacity 9 times greater, with mass production expected in 2027 [1] - Tesla signed a $16.5 billion agreement with Samsung for the production of the AI6 chip in the U.S., anticipated to launch in 2028 [1] Group 2 - Anthropic has upgraded Claude Cowork with a "permanent memory" feature, allowing the AI to categorize and store information, enhancing user understanding over time [2] - The upgrade includes an MCP connector system to improve automation, voice mode development, and a new UI area for continuous management of results [2] - Continuous learning is viewed as a key breakthrough for AGI, with OpenAI and Google also investing in memory functionalities [2] Group 3 - Kunlun Wanwei has launched Skywork Design Agent, focusing on poster design, social media materials, logo branding, and general creative image generation [3] - The product features a self-developed canvas engine that supports manual editing, AI photo retouching, and layer separation, streamlining the entire process from material import to export [3] - It offers multiple export formats (PNG, JPG, PDF) and includes a unique "add to knowledge base" feature to address material management issues, now fully launched overseas [3] Group 4 - Douzi 2.0 has introduced the Coze Skill feature, allowing users to encapsulate personal methodologies and industry experiences into reusable "skill packages" [4] - A new "long-term plan" feature enables goal-oriented AI collaboration, breaking down vague objectives into clear steps for automatic execution [4] - The launch of a skill marketplace facilitates the exchange of industry skill packages, allowing professionals to monetize their expertise, alongside the beta release of video Agent Skill [4] Group 5 - Giant Network's "Supernatural Action Group" has introduced an "AI large model challenge" mode, integrating large model technology into game combat, marking a significant application in a high DAU game [5] - AI characters are driven by large models in real-time, capable of voice interaction and mimicking human behavior, with over 25 million AI matches recorded in the first week [5] Group 6 - Anker and Feishu have collaborated to create a 10-gram AI recording device, addressing the portability issues of traditional AI recording cards [7] - The device features real-time summarization capabilities, generating structured logical maps during meetings and supporting real-time translation in 24 languages [7] - Recordings are directly streamed to the Feishu knowledge base, integrating with the entire Feishu ecosystem to reduce the burden of knowledge base construction [7] Group 7 - Roboto has open-sourced its bipedal humanoid robot prototype, achieving a running speed of 3 m/s, making it one of the most advanced open-source humanoid robots globally [8] - The open-source content includes hardware schematics, EBOM material lists, supplier information, and control algorithm code, enabling reproducibility and verification [8] - The team, originating from Harbin Institute of Technology, has secured millions in seed funding and aims to reduce the cost of embodied intelligence development by 80% [8] Group 8 - Galaxy General has launched the Galbot S1, a heavy-duty robot capable of carrying loads up to 50 kg, currently operating in key production processes at CATL [9] - It features an industry-first fully autonomous, zero-remote operation "embodied handling model," utilizing pure visual perception without QR code markers [9] - Galaxy General has recently completed a 2.1 billion yuan financing round, with a valuation exceeding 20 billion yuan, and has established partnerships with leading manufacturers [9] Group 9 - OpenAI's product manager reported that since the release of ChatGPT-5, the Codex platform has seen a 20-fold growth, processing trillions of characters weekly [10] - The Sora Android app achieved a rapid development cycle, going from zero to launch in 28 days and topping the App Store, significantly improving team efficiency [10] - The manager noted that human typing speed and multitasking capabilities are often the limiting factors for AGI, rather than the models themselves [10]