Meta Plans to Launch Its “Hatch” AI Agent Platform in the Coming Weeks
On August 24, reports surfaced that Meta Platforms is aiming to introduce its consumer-facing OpenClaw AI assistant (internally codenamed “Hatch”) as early as the next few weeks, with its latest AI model, “Watermelon,” also slated for an October release.
8.86 Seconds! Humanoid Robot Sprint Record Shattered Again
On August 25, during the second World Humanoid Robot Games, the 100-meter repechage for the large-sized group saw a brand-new record. Following the 9.39 seconds clocked at the opening ceremony, the Tiangong robot from the Beijing Humanoid Robot Innovation Center blazed down the track in an impressive 8.86 seconds, once again resetting the benchmark.
NVIDIA Announces Groq 3 LPX Is Now in Full Production
NVIDIA announced on August 25 that its interactive AI inference accelerator, the NVIDIA Groq 3 LPX, has officially entered mass production. As an extension of the NVIDIA Vera Rubin platform, the Groq 3 LPX significantly boosts AI reasoning capabilities by enabling lightning-fast token generation for highly responsive agentic systems.
Tencent Restructuring Its AI Gaming Division? Company Insider Sets the Record Straight
According to a report from The Beijing News’贝壳财经, in response to speculation about Tencent’s Interactive Entertainment Group (IEG) reorganizing its AI gaming department, a Tencent representative clarified that the article contains substantial inaccuracies, and the so-called “creation of a new division” is a misinterpretation. Recent rumors suggest that IEG is consolidating its ARC Lab under the CROS (Tencent Gaming Technology Middle Platform) with its Gaming AI Engine Department, with plans to expand the team to 400-500 people. The aforementioned insider also noted that the gaming technology shared platform has long housed a Gaming AI Engine Department, which has been actively recruiting AI specialists. This team is dedicated to enhancing AI-driven efficiency across the game development pipeline and fostering gameplay innovation, with the goal of providing reusable technical capabilities and solutions for game projects.
ByteDance Launches Doubao Work, Deep Integration with Feishu Builds Enterprise-Grade Agents
On August 25, Doubao Work was officially unveiled. As a brand-new Agent product and brand from ByteDance aimed at productivity scenarios, Doubao Work can autonomously break down tasks around user goals, invoke tools, and drive complex workflows forward. It also integrates deeply with Feishu, enabling Agents to complete tasks with greater precision by tapping into enterprise context. Currently, the desktop version of Doubao Work is available for download via the official website, and users can also access it directly in the latest version of Doubao for PC.
Robam’s Next-Gen AI Cooking Glasses Expected to Launch Next March
According to a report from Lanjinger News, a source familiar with the matter revealed that Robam (老板电器) is currently developing its next-generation AI cooking glasses, which are expected to be officially released in March next year. Unlike the first-generation product, which was only showcased for offline experience, this new version will actually be sold directly to consumers.
Guangxiang Technology Unveils Physics-Native World Model ActEffect, Giving Embodied AI a True Grasp of Physical Causality
On August 25, it was reported that Guangxiang Technology has released its physics-native world model, Phi-WM 1.0 ActEffect. This marks the first concrete model-based achievement of its “Physics-native Intelligence” (Phi) technical roadmap.
Unlike traditional VLA models that rely on imitation learning or generative world models built on statistical prediction, ActEffect transforms the world model into a “training-time feedback mechanism.” This enables robots to genuinely establish sequential causal relationships between “actions and their physical consequences.” Since causal learning is completed during the training phase, there’s no need to retain the world model for explicit iterative predictions during inference. As a result, it achieves lightweight reasoning and significantly cuts computational costs.
