OpenAI Official Updates

Boston Children’s uses AI to unlock new diagnoses

OpenAI

Boston Children’s Hospital has deployed OpenAI’s artificial intelligence technology in clinical scenarios, with core value reflected in two key areas: First, it optimizes the quality of patient care services, while reducing the hospital’s operational burden and improving overall operating efficiency. Second, the technology has assisted clinicians in successfully diagnosing more than 40 rare disease cases, providing a feasible technical solution to the clinical pain points of difficult rare disease diagnosis and high misdiagnosis rates, and also offering a reference for AI implementation in pediatric diagnosis and treatment scenarios.

How Braintrust turns customer requests into code with Codex

OpenAI

This practice article reveals the R&D efficiency improvement solution of the Braintrust team: Engineers have integrated the code generation model Codex with GPT-5.5 into the development process, enabling efficient conversion of customer requirements into runnable code, while greatly shortening the verification cycle for technical experiments. Overall code production efficiency and customer demand response speed have been significantly improved, providing an implementation reference for large models empowering the R&D pipeline.

Anthropic News

Introducing Claude Opus 4.8

Anthropic

Claude Opus 4.8 is the latest iteration of the Opus-level large model. The core capabilities of this version have been upgraded across multiple dimensions: processing performance in three scenarios, namely programming development, agent tasks, and professional field work, has been significantly improved, while operational stability for long-cycle runs has been optimized to adapt to more complex continuous work requirements, further enhancing its comprehensive practical value.

Introducing Claude Design by Anthropic Labs

Anthropic

Anthropic Labs has officially released its new product Claude Design. The product supports users to collaborate with the Claude large model for creation, jointly producing highly mature visual outputs covering multiple common visual work scenarios such as graphic design, product prototypes, presentation slides, and single-page promotional materials, further expanding the application boundaries of generative AI.

Google DeepMind

We’re launching the Google DeepMind Accelerator program in Asia Pacific to tackle environmental risks

Google DeepMind

Google DeepMind officially launches its Asia Pacific Accelerator program, with the core goal of addressing various regional environmental risks with cutting-edge AI technology. The program will provide technical empowerment and resource matching support for innovative teams and research entities in the environmental field, focusing on promoting the implementation of AI solutions in scenarios such as climate response, disaster early warning, and ecological protection, to help improve the overall effectiveness of environmental risk prevention and control in the Asia Pacific region.

Fast-tracking genetic leads to reverse cellular aging

Google DeepMind

This study focuses on the discovery of genetic targets for cellular aging reversal. Biologists conducted screening using the Co-Scientist research tool, and successfully identified a previously unreported new regulatory factor, which was verified to effectively achieve human cell rejuvenation. This achievement greatly shortens the R&D cycle of anti-aging targets, providing a new candidate direction for subsequent clinical intervention of aging-related diseases and anti-aging drug development.

Hugging Face Blog

Profiling in PyTorch (Part 1): A Beginner’s Guide to torch.profiler

Hugging Face

This is an introductory guide to torch.profiler for PyTorch users, focusing on PyTorch’s built-in performance analysis tool, covering full entry-level operations including basic tool configuration, commonly used calling APIs, and performance report interpretation. It helps beginners quickly master the method of locating performance bottlenecks in model training and inference phases, troubleshoot problems such as abnormal operator time consumption, low GPU utilization, and excessive memory usage, to optimize model operating efficiency.

ITBench-AA: Frontier Models Score Below 50% on the First Benchmark for Agentic Enterprise IT Tasks — by Artificial Analysis and IBM

Hugging Face

Artificial Analysis, in partnership with IBM, released ITBench-AA, the world’s first special evaluation benchmark for agents targeting enterprise IT scenarios, which specifically assesses the ability of large model agents to handle professional enterprise IT tasks such as O&M deployment and troubleshooting. Actual tests show that current leading global large models score less than 50% on this benchmark, indicating that there are still significant capability gaps for large models to be deployed in enterprise-level IT agent scenarios.

The Gradient

After Orthogonality: Virtue-Ethical Agency and AI Alignment

The Gradient

This AI alignment research based on virtue ethics challenges the premise that “rational agents must be bound to fixed ultimate goals”, proposing that human rationality is not directed at specific goals, but rather that actions adapt to social practice networks that include rules and evaluation systems. It argues that if AI is to adapt to and collaborate with humans, its decision-making logic must match the human practice-oriented paradigm, which is critical for ethical alignment and core security guarantees.

QbitAI

AI原生时代下，让世界适应Agent，而非教AI做人 | 港大黄超@AIGC2026

QbitAI

Huang Chao, Assistant Professor at the University of Hong Kong, put forward the core idea of the Agent era at the 2026 China AIGC Industry Summit: Instead of making Agents adapt to humans, we should transform the digital world to adapt to Agents. The lightweight general-purpose Agent nanobot open-sourced by his team has exceeded 200,000 downloads, receiving high recognition from the industry. Next, the team will tackle cross-ecosystem long-range tasks, and has also launched the CLI-Anything solution, which transforms professional software into command-line interfaces compatible with Agents, reconstructs interaction paradigms, and promotes the upgrade of Agents into digital labor.

从Token无上限到全员Agent：MiniMax的AI Native组织进化实践

QbitAI

At the 2026 China AIGC Industry Summit, MiniMax, a multimodal large model enterprise that just listed on the Hong Kong Stock Exchange, shared its AI-native organization practices: It is recommended to start with high-value scenarios that employees are initially resistant to, distribute Token subsidies to encourage employees to use Agents to build automated workflows, and use Token consumption as a new efficiency indicator, which can promote organizational flattening. The company judges that AI will deeply penetrate all industries in the next 2-3 years, and enterprises do not need to be anxious, they can just start practicing directly.

帮Gemini拿下IMO金牌的关键先生，差点成了职业钢琴家

QbitAI

This article introduces Yi Tay, a research scientist at Google DeepMind and alumnus of Nanyang Technological University: He almost became a professional pianist, and his performance of Chopin’s Fantaisie-Impromptu 14 years ago was highly skillful and full of emotional tension. As a core member, he led the team that developed Gemini Deep Think which won the IMO gold medal, and he is also a core contributor to Gemini 2.5 and 3 Deep Think, the latter of which also reached gold medal level in the written test of the International Physics and Chemistry Olympiad.

---