OpenAI Official Updates

OpenAI named a Leader in enterprise coding agents by Gartner

OpenAI

Gartner has just released its 2026 Magic Quadrant report for enterprise-grade AI coding agents, where OpenAI was awarded the highest rating of Leader, placing it among the top vendors in the track. In this evaluation, the technical innovation of OpenAI’s Codex large coding model and its large-scale deployment capability adapted to enterprise scenarios were highly recognized by reviewers, which served as the core support for it to obtain this rating.

How Virgin Atlantic ships faster with Codex

OpenAI

In its mobile app revamp project, Virgin Atlantic adopted the AI programming tool Codex to improve R&D efficiency, and successfully launched the product before the strict delivery deadline ahead of the fixed holiday travel season. The project not only achieved nearly 100% unit test coverage, but also had no P1 (highest priority) defects in the final released version, fully verifying the dual positive effects of AI coding tools on both project delivery efficiency and output quality.

Anthropic News

Introducing Claude Opus 4.7

Anthropic

The latest large model Claude Opus 4.7 has been officially opened for general use recently. Compared with the previous generation Opus 4.6, the core upgrade of this model focuses on the field of advanced software engineering, with significantly improved overall performance in corresponding scenarios, especially a more prominent breakthrough in the processing capability of high-difficulty software engineering tasks, achieving considerable iterative upgrade of overall performance.

Introducing Claude Design by Anthropic Labs

Anthropic

Anthropic Labs has newly launched the collaborative AIGC product Claude Design, which focuses on the human-machine collaborative creation mode: users can work with the Claude large model to produce various types of mature and standardized visual works such as design drafts, interactive prototypes, presentation slides, and single-page promotional materials. This product complements the implementation application scenarios of Anthropic’s large models, and also lowers the threshold for producing high-quality visual content for non-professional design users.

Google DeepMind

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

Google DeepMind

Google DeepMind recently officially launched its Asia Pacific accelerator program, which focuses on regional environmental risk governance. The program will integrate its own AI technical advantages and industrial ecosystem resources to support regional scientific and technological innovation teams and environmental research institutions, focus on Asia-Pacific specific environmental scenarios such as floods, extreme high temperatures, and biodiversity degradation, and implement AI prediction and governance solutions to effectively improve the intelligent response efficiency of regional environmental risks.

Fast-tracking genetic leads to reverse cellular aging

Google DeepMind

Biology researchers focused on the demand for efficient mining of genetic targets for cellular senescence reversal, carried out screening relying on the intelligent scientific research assistance tool named Co-Scientist, and successfully identified unreported new regulatory factors. Experiments have verified that these factors can effectively achieve the rejuvenation of human cells, providing new candidate targets and research ideas for anti-aging technology research and development and intervention of aging-related diseases.

Hugging Face Blog

Towards Speed-of-Light Text Generation with Nemotron-Labs Diffusion Language Models

Hugging Face

NVIDIA’s Nemotron-Labs team has launched a new diffusion language model, targeting the efficiency pain point of word-by-word inference of traditional autoregressive text generation, optimizing the diffusion iteration step size and sampling strategy, achieving text generation efficiency approaching “speed of light”, which is several times faster than mainstream autoregressive large models, while ensuring the semantic accuracy and fluency of generated text, and can adapt to the needs of high real-time interactive scenarios.

Specialization Beats Scale: A Strategic Variable Most AI Procurement Decisions Overlook

Hugging Face

This paper focuses on the common blind spot in AI procurement decisions, pointing out that the current industry mostly prioritizes large-parameter general models, but ignores the core variable of scenario specialization. The research verifies that vertical small models customized for specific businesses are better than large general models called in the same scenario in terms of accuracy, deployment cost, data security and response efficiency. It is recommended that procurement take scenario adaptability as a core evaluation indicator, and there is no need to blindly pursue model scale.

The Gradient

After Orthogonality: Virtue-Ethical Agency and AI Alignment

The Gradient

This paper conducts AI alignment research from the perspective of virtue ethics, challenging the underlying presupposition of the orthogonality hypothesis that “rational agents are necessarily bound to fixed goals”, pointing out that human rationality is not directed to the ultimate goal, but adjusts behavior in the practice network composed of actions, tendencies, evaluation criteria, etc. The study proposes that if AI is to adapt to collaborating with humans, its decision-making logic needs to be isomorphic with human practice-driven logic, which is not only an ethical alignment requirement, but also related to the core security of AI.

QbitAI

”Five Types of People Cannot Be Replaced by AI, It Is Safest for Enterprises to Be the Second Runner” | Fang Han of Kunlun Wanwei @ AIGC2026

QbitAI

At the 2026 China AIGC Industry Summit, Fang Han, CEO of Kunlun Wanwei, put forward core judgments: Token is the electricity of the AI era, and its consumption has become a new hard power indicator of the industry; closed-loop fault-tolerant positions are easily replaced by AI, only positions with high judgment requirements and non-fault-tolerant attributes are stable, and there are five types of people who will not be replaced by AI; AI will dissolve the middle layer of the workplace, leaving only novices and top experts; most of the remarks advocating that AGI is coming are to boost market value, and it is more stable for enterprises to be the second in the industry.

OpenAI Expert Teaches You How to Make the Most of Codex

QbitAI

Jason Liu, a new member of the OpenAI Codex team and author of the 13k-star open source project Instructor, has publicly shared advanced gameplay of Codex: different from the habit of ordinary users closing the session after a single use, the core is to retain long-life cycle threads to accumulate project context, use voice input instead of typing instructions, and match the Heartbeats + @computer function to transform Codex into a long-term work system that can independently handle refunds and receive requirements. Combined with recent Codex updates such as direct screenshot input and official launch of Goal mode, its agency work capability has been greatly improved, and its weekly active users have now exceeded 4 million.

DeepSeek V4 Prices Slashed Sharply, CATL, JD.com, NetEase Rush to Invest, Liang Wenfeng: The Goal is AGI

QbitAI

DeepSeek recently announced two developments: first, the V4-Pro API will be permanently reduced to the current promotional price; second, it is promoting a 70 billion yuan financing with a valuation of 45 billion US dollars, and is negotiating investment with CATL, which is laying out data center energy storage business, as well as JD.com and NetEase. Founder Liang Wenfeng made it clear that the company will give priority to tackling AGI, adhere to open source, and not pursue short-term commercial monetization.

---