01News
OpenAI announces a pilot program for the Grid Protectors initiative at its Effingham County, Georgia data center. It pledges to release up to 1,000 MW of power during extreme weather or emergencies, enough to power air conditioning for approximately 700,000 households.
Source links can be verified on the original siteRead original ↗02News
The New York Times joins forces with New York Daily News, Chicago Tribune, Orlando Sentinel, and other publishers to sue Brave Software.
Source links can be verified on the original siteRead original ↗03News
Skyfall AI has released SKY-001, which can design physical products in 3D by parts and construction steps. Its CAD precision matches GPT-6 Astra, with a single design cost approximately 150 times lower. The model outputs part trees, operation sketches for each step, and intermediate entities, which are decomposed and assembled by Hierarchical Composer and predicted by Construction World Model in two stages.
Source links can be verified on the original siteRead original ↗04News
OpenAI has released LifeVal, a benchmark testing whether AI models can complete everyday tasks for people. GPT-6 Astra scored 80.6% on daily tasks, compared to 49.3% for GPT-5.2 less than nine months ago. LifeVal covers caregiving, health, education, finance, shopping, and household administration, ranging from booking vehicle inspections to ordering halal grocery items within a budget.
Source links can be verified on the original siteRead original ↗05News
A new Meta paper finds that when AI agents learn from past attempts, matching new problems to notes of similar issues performs better than sharing a single full history. The proposed Routed CL uses an LLM router to reuse or create memory clusters before each run, with each cluster employing one of five memory types (such as raw historical runs or distilled experience).
Source links can be verified on the original siteRead original ↗06News
DeepSeek is promoting its free AI agent app, Harness. Users can describe desired tools in natural language, and the agent will write, install, and test them as new in-app plugins.
Source links can be verified on the original siteRead original ↗07News
Google summarizes key October 2026 AI developments. The focus is on the deployment of the reasoning model Gemini 4 Argon globally, across Google AI Studio, the Gemini API, Google Cloud, Search, and the Gemini app.
Source links can be verified on the original siteRead original ↗08News
OpenRouter reveals its weekly token processing growth trajectory: it took 50 weeks to rise from 1 trillion to 10 trillion tokens, and only 28 more weeks to reach 100 trillion. The platform now states that a weekly volume of 1 quadrillion (one quintillion) is approaching.
Source links can be verified on the original siteRead original ↗09News
LangChain notes that as active, persistent, and goal-oriented agents rise, context engineering faces a new set of challenges. The article lists issues such as balancing proactive pushes with interruptions, memory trade-offs, asynchronous sub-task notifications, goal tracking, goal alignment (such as deleting the database when instructed to 'fix all bugs'), and browser tool context management."Eliminate all bugs"issues such as database deletion and browser tool context management, noting that no universal solution currently exists in the industry.
Source links can be verified on the original siteRead original ↗10News
What are the new features of Grok Bot? 00:14 Shopping Cards 00:27 Product Pages 00:39 Order Tracking 00:53 Payment 01:33 Custom Modes 01:52 Team Bot Login and Settings 02:09 Sharing and Finding Team Bots 02:51 Connector requests in Slack 03:12 Contacts in Vault 03:31 Quick Q&A
Source links can be verified on the original siteRead original ↗11News
OpenRouter data analysis shows that users consume more models from multiple providers rather than sticking to one family. October 2026 figures indicate that customers spending between $100 and $1,000 monthly use an average of five labs; those spending $1,000 to $10,000 use eight; and those spending over $10,000 use twelve. Among high-spending customers ($10K+), 97% use three or more labs, while 90% use five or more. For customers spending over $100 monthly, less than 4% of users rely on a single lab exclusively; even the most sticky provider, Anthropic, only accounts for 39% of its own customers' spending.
Source links can be verified on the original siteRead original ↗12News
Following this shift, the number of model labs per customer paying over $100 monthly has risen each quarter. By October 2026, it reached 7.3, the highest value we have measured to date. The median increased from 3 to 6, and the top 10% grew from 7 to 14. Overall, the figure roughly doubled.
Source links can be verified on the original siteRead original ↗13News
Companies and users on OpenRouter spending more than $100 monthly now use an average of six model families, double the number from last year. A thread of tweets and videos about model neural diversity and why it is accelerating 👇
Source links can be verified on the original siteRead original ↗14News
According to Bloomberg and The Information, Anthropic's revenue for the quarter was nearly $17 billion, twelve times its figure from a year ago. However, operating costs reached $17.5 billion, resulting in a quarterly net loss of $23.3 billion, which exceeds the $12.1 billion recorded in the same period last year.
Source links can be verified on the original siteRead original ↗15News
Israeli startup 42AI has come out of the shadows after raising $62.5 million in seed funding. The company is developing a "behavioral world model" designed to predict human decision-making. This model has been deployed at approximately 10 major investment banks to forecast how traders react during volatile markets. 42AI claims its prediction accuracy for trading direction achieves an F1 score of 0.8 to 0.85. The startup states the model does not require personal identifiable information and plans to expand applications into gaming, fraud prevention, and cybersecurity.
Source links can be verified on the original siteRead original ↗16News
Musk confirmed on X that the weekend Grok Bot usage limits being reached by some users were not a glitch. He cited excessively high service costs relative to revenue as the cause and did not explain how subscriber caps will be adjusted in the future. The article also notes that SpaceX disclosed last month that AI accounted for 52% of its revenue, and its renamed Super Intelligence division reported $431 million in profit. However, this figure was lower than potential earnings due to a one-time $1.2 billion cost from acquiring Cursor.
Source links can be verified on the original siteRead original ↗17News
Anthropic updates its S-1 IPO filing and reveals Q3 performance for the first time. Operating profit turned from $974 million in Q2 to a $567 million loss in Q3, primarily due to $5 billion in high-cost short-term compute and $2 billion in non-cash employee donation matching.
Source links can be verified on the original siteRead original ↗18News
Meta's Agent Arena has added a performance comparison view, allowing users to compare how different agents perform per token output and per minute of wait time. Real-world usage data shows that under equal intelligence levels, costs can differ by up to 10 times, token output can be six times higher, and completion time can be twice as long. Additionally, the number of tokens generated is almost impossible to predict from the bill.
Source links can be verified on the original siteRead original ↗19News
Sam Altman said at the opening ceremony of ADIPEC in Abu Dhabi that early this year most AI systems were not as good as the world's top human scientists, but now they mostly are. He also noted that while AI could not previously handle many white-collar jobs, it can now complete many of them. He emphasized that this assessment is rough and does not hold true in all aspects.
Source links can be verified on the original siteRead original ↗