News

On October 23, 2026, Pathos AI announced the completion of a $237 million Series E funding round, bringing its post-investment valuation to approximately $3 billion. The round was led by Bristol Myers Squibb (BMS) with participation from existing investors. The funds will be used to expand F...

On October 22, 2026, Microsoft announced that a study on the diagnostic system MAI-DxO and the sequential diagnostic benchmark SDBench had been officially published in *Science*. Early results from this project had already been released in 2025; what is noteworthy this time is the expanded evaluation, public release...

A report on a high-risk vulnerability was shared in a group chat, but there is still a long way to go before production systems are truly secure—including an asset inventory, vendor updates, compatibility testing, deployment, and operational verification. This PHP security update has once again brought this often-overlooked process to the forefront for enterprises. On October 21, PHP...

It’s not hard to get an AI to find a carton of milk. But getting it to choose the right size, use the correct payment method, know when to stop when prices change, and provide a clear resolution process in case of a mistake—that’s what comes close to a sustainable business. On October 21, Walmart announced its arrangements for shopping agents: A...

More than 330,000 Volkswagen Group employees have been equipped with Copilot, marking a milestone in the phased rollout rather than a one-day deployment. This article distinguishes between employee coverage, monthly active users, and active agents; analyzes work context, data permissions, and organizational governance; and discusses how enterprises can verify actual returns.

NVIDIA is expanding the Agent Toolkit to connect search, business semantics, structured data forecasting, and optimization computing. Enterprise AI must not only find the right data, but also standardize metrics, preserve source and access boundaries, and assess the value of implementation based on a comprehensive cost analysis.

A large-scale steel model capable of answering technical questions and generating analytical reports can easily capture the audience’s interest during a demonstration. But once the procurement process begins in earnest, the questions become more specific: Can the system still function if a different steel grade is used? Are the recommendations still reliable if the conditions of raw materials or equipment change? Who takes responsibility if something goes wrong? If these issues cannot be repeatedly verified...

OpenGCUA: An Open Execution Stack for Computer Agents. This article analyzes how enterprises can assess the true value of computer agents by examining models and frameworks, evaluation metrics, implementation costs, permission boundaries, and pilot methods.

A batch of production materials has encountered issues, and the impact may extend beyond the production department. Whether follow-up research needs to be adjusted, whether clinical trials need to be postponed, and which partners should have their schedules rescheduled may all depend on this single piece of news. Challenges in drug development arise not only at the lab bench but also at the interfaces between departments and projects...

Whether an AI chip makes it onto a company’s procurement list often depends on more than just computing power and price. Engineering teams are more concerned with another issue: once existing models are migrated, how much code will need to be rewritten, how long will it take for them to run stably, and who will resolve any issues that arise? On October 19, 2026, PyTorch...

On October 19, 2026, Inception released Mercury 3. The official key metrics are 1,068 tokens per second and a 1M-token context, and the company claims that in a set of benchmarks, it approaches or exceeds performance-optimized...

It’s certainly appealing to be able to “schedule a review meeting” and “find the conclusions from the last discussion” within the same AI chat window. Tencent Meeting MCP/Skill provides this integration: the assistant can manage meetings and access recordings and meeting minutes. The resulting change is that by using...

On October 19, 2026, MiniMax officially released M3.1 Flash, announcing that the total cost of the Agentic task had been reduced by 70% compared to the previous generation, while online inference speed had increased nearly threefold. The API, MPlan, and MiniMax Cod...

Model providers such as Zhipu and MiniMax have launched stores on Tmall, and the AI Token recharge center has entered the Double 11 shopping festival. What does the retailization of AI services mean? From billing units and package boundaries to renewal and performance, this article breaks down the new logic behind consumer decision-making.

On October 17, DeepSeek released an update to V4.1 Flash and made the Search and Fetch APIs available. Model pricing remains unchanged, but tool calls are billed separately. How should developers evaluate performance, search quality, and the total cost of a task?

The Ministry of Commerce and the Ministry of Science and Technology are soliciting public comments on revisions to the Technology Export Catalog and are simultaneously implementing a temporary pre-export registration system for relevant technologies. How should AI companies distinguish between policy phases, technology delivery, and service invocation, and reassess their overseas expansion strategies and partnership boundaries?

ChatGPT Slides is gradually rolling out real-time collaborative editing and shared comments. This article analyzes the practical value of team collaboration, the limitations of web and desktop exports, and the issues related to permissions, content quality, and cross-software delivery that enterprises should verify when evaluating the service.

An internal study by Anthropic on employment reveals that the expansion of AI agents is transforming recruitment and the organization of work. This article breaks down the distinctions between recruitment share, agent working hours, and productivity, and discusses the burden of vetting, onboarding new hires, and actionable improvements for companies.

RTX Spark brings local AI, CUDA, and Windows to thin-and-light laptops. The top-of-the-line configuration and the starting price can’t be combined into a single machine; the real factors to consider when making a purchase are still model speed, software compatibility, sustained workload, and frequency of use.

GitHub has revamped its key detection by combining rule-based filtering with context-based classification. Cost savings exceeding 98% fall within a clearly defined range; what’s even more noteworthy is how smaller models are changing the timing of security checks, and how enterprises measure the total cost of false positives, false negatives, and true positives.