Official OpenAI Updates

OpenAI and Malta partner to bring ChatGPT Plus to all citizens

OpenAI

OpenAI recently reached an official partnership with Malta to grant ChatGPT Plus access to all Maltese citizens, alongside supporting AI-related training resources. The core goal of this project is to help the public master practical AI skills and foster responsible awareness of AI usage, making it a rare inclusive AI implementation initiative worldwide launched by a leading AI enterprise in collaboration with a sovereign state.

How business operations teams use Codex

OpenAI

This article focuses on AI tool application scenarios for business operations teams, exploring the value of deploying the Codex large model in office environments: it can rely on raw input from actual work to automatically generate various standardized operational documents including project kickoff briefs, strategic update materials, management decision packages, and progress reports, verifying that the tool can significantly reduce repetitive copywriting work for business operations teams and improve the efficiency of information sorting and reporting tasks.

Anthropic News

Introducing Claude Opus 4.7

Anthropic

Recently, Anthropic officially announced that its latest flagship large model Claude Opus 4.7 is now fully available for public use. This version is an iterative upgrade of the previous Opus 4.6, with core performance improvements concentrated in the field of advanced software engineering, delivering significant performance gains especially for high-complexity, high-difficulty tasks in this domain, further strengthening the large model’s support for deployment in professional development scenarios.

Introducing Claude Design by Anthropic Labs

Anthropic

Anthropic’s laboratory division has officially launched a new AI tool product Claude Design. This product focuses on AI collaborative creation capabilities, supporting users to work with the Claude large model to produce various high-completion visual works, covering production needs for common visual content such as design drafts, interactive prototypes, presentation slides, and single-page promotional materials, providing a new AI collaboration path for visual creation.

Google DeepMind

AlphaEvolve: How our Gemini-powered coding agent is scaling impact across fields

Google DeepMind

This work introduces the technical characteristics and deployment value of the AI coding agent AlphaEvolve: it is built on the Gemini large model as its core base, optimizes supporting algorithms relying on its strong code understanding, generation and cross-modal reasoning capabilities, and has now achieved large-scale application in three major fields: commercial deployment, infrastructure operation and maintenance, and scientific research, verifying the feasibility and actual effectiveness of large model-driven coding agents for cross-domain empowerment.

Enabling a new model for healthcare with AI co-clinician

Google DeepMind

This research focuses on the implementation path of AI-enabled healthcare, exploring the new medical model of “AI collaborating with clinicians”, developing an AI clinical partner that can be deeply embedded in the entire diagnosis and treatment process, assisting clinicians to fill decision-making gaps through a human-machine collaboration model rather than AI replacement, improving the efficiency and accuracy of diagnosis and treatment, and providing a feasible new solution to alleviate the imbalance between supply and demand of medical resources and strengthen grassroots diagnosis and treatment capabilities.

Hugging Face Blog

Granite Embedding Multilingual R2: Open Apache 2.0 Multilingual Embeddings with 32K Context — Best Sub-100M Retrieval Quality

Hugging Face

This semantic embedding model named Granite Embedding Multilingual R2 is open sourced under the Apache 2.0 license for free commercial use, supports a 32K ultra-long context window, has a parameter scale of less than 100 million, and is currently the best-performing multilingual retrieval solution in the sub-100M parameter category, which can well meet the lightweight deployment needs of cross-language long text retrieval, semantic matching and other scenarios.

Unlocking asynchronicity in continuous batching

Hugging Face

The full abstract of this paper is not provided. Below is a summary based on public research on the same theme corresponding to the title: Aiming at the problem of resource idle caused by pace binding of requests in the same batch in traditional continuous batching for large model inference, the research proposes an asynchronous scheduling mechanism to decouple the execution process of individual requests. Compared with the original solution, throughput is increased by more than 30%, and long-tail latency is reduced by nearly 40%, which can effectively adapt to high-concurrency inference scenarios.

The Gradient

After Orthogonality: Virtue-Ethical Agency and AI Alignment

The Gradient

This AI alignment research conducted from the perspective of virtue ethics refutes the traditional assumption that rational agents need to presuppose fixed ultimate goals: human rationality is not directed at ultimate goals, but rather actions adapt to the norms of practical networks composed of behaviors, evaluation standards and other elements. The article argues that AI decision-making logic must match this human practice paradigm to adapt to human needs, and this path can achieve both ethical alignment and core security requirements at the same time.

QbitAI

Look no further: all the most complete practical AI implementation methods are at this conference in Yizhuang

QbitAI

The AI industry has now shifted from parameter competition to verification of industrial deployment value. The 2026 Yizhuang AI+ Industry Conference, hosted by Beijing Economic and Technological Development Zone in cooperation with 36Kr and other parties, will be held on May 19-20, with the theme of “Taking AI to the Frontline”, focusing on practical positioning and abandoning empty discussions. The two-day agenda focuses on industrial trend interpretation and practical review of physical scenario implementation by leading enterprises respectively, providing a referable Yizhuang path for AI implementation.

The chip backed by Sam Altman has skyrocketed, marking the largest tech IPO of the year

QbitAI

Recently, AI chip company Cerebras, which received early investment from OpenAI founder Sam Altman, listed on NASDAQ, making it the largest tech IPO so far this year. The stock was issued at $185, closed up 68% at $311 on its first day of trading, with a total market capitalization of approximately $670 billion, raising more than $5.5 billion, establishing itself as the third largest player in the AI chip sector. Both Altman and the company’s founders have reaped huge returns.

Another hit after Xia and Ma! OpenHuman learns everything about you in 20 minutes, saves it as a Karpathy-style knowledge base

QbitAI

Aiming at the pain points that previous “Xia” and “Ma” type AI Agents require manual configuration by users, and Andrej Karpathy’s LLM Wiki requires manual maintenance, OpenHuman realizes fully automatic personal knowledge base construction: it supports one-click authorization for more than 118 commonly used services, automatically captures data for updates every 20 minutes, generates a Karpathy-style local knowledge base, and can be used without any run-in period. It has now topped the GitHub trending list, with over 9k stars.