JetBrains open-sources Mellum 2.1, which outperformed Qwen 3.5-9B in LiveCodeBench tests.

Insight Beating AI News: 9-year-old Yu Zhile, a fourth-grade pupil from Xiamen, Fujian, has recently established the "Zhile AI Scholarship" at Longshan Middle School in Nanjing, Zhangzhou, to award scholarships to students at the school. According to a report by *Minnan Daily*, Yu started learning AI in January this year, then used AI tools to develop products and websites and attempted to sell them. Including earnings from competition awards, his cumulative income is reportedly six-figure RMB. His mother Zhuang Sehui runs an internet company in Xiamen, and Longshan Middle School is her junior high alma mater. She has long supported students in her hometown. This time, the mother and son decided to set up the scholarship with part of their income, planning to reward outstanding students annually. Longshan Middle School Principal Wu Ruixing stated that the scholarship will be used to encourage students who excel in both character and academics and have a passion for technological innovation.

Beating AI Express News reports that Pine AI has launched Pine Computer, a cloud computer designed specifically for AI Agents. It integrates cloud virtual machines, AI models, and operational tools. Developers submit tasks via API, enabling the AI to operate browsers, software, and files in the cloud without requiring users to use their own devices. Pine’s core improvement is enabling the computer to proactively inform the AI of changes, addressing a key limitation of traditional vision-based Agents, which typically need repeated screenshots to assess click outcomes. Pine instead lets browsers and operating systems directly feed back changes to the AI model executing the task, cutting down on repeated screenshots and page checks. Pine claims its new solution completes tasks 2 to 5 times faster than traditional approaches in internal tests. In the SaaS-Bench v1.1 test, Pine’s checkpoint score hit 78.3%, outperforming Claude Code’s 74.3% and Codex’s 71.1%. This score measures how many check requirements are fulfilled during task execution, not overall task success rate. However, X Community notes point out Pine’s full task completion rate is only 27.4%, lower than Claude Code’s 31.1% and Codex’s 29.2%. Pine’s model costs are also far lower, averaging around $1.02 per task, versus $20.50 for Codex and $26.50 for Claude Code. That said, these fees exclude cloud server costs, and each solution uses different models and execution budgets. Currently, Pine Computer remains in an invitation-only closed beta phase.

Beating AI Express (Dongcha Insight) news: Anthropic’s July 23 updated transparency report shows it banned 11.4 million accounts in the first half of 2026. The figure for the second half of 2025 was 1.45 million, meaning the latest number is roughly 7.9 times that of the prior period. The statistics cover all of Anthropic’s products and services, with no separate breakdown for Claude chat and developer offerings. During the same period, the company received 398,000 account ban appeals, of which 42,000 overturned the original ban decision. Calculated based on the two same-period data, overturned cases account for approximately 10.6% of total appeals. Anthropic noted that accounts may be banned for violating usage policies, terms of service, or supported region policies.

Robinhood ecosystem AI project orbio has announced a $1.2 million seed funding round, with participating institutions undisclosed. The project’s official statement says the funds will be used to build rails enabling agents to self-configure compute, hardware, and top-tier private infrastructure. As of press time, GMGN data shows ORBIO’s token market cap stands at $5.342 million, with a 24-hour increase of 26%. ORBIO is issued on the Pons platform, with Nvidia stock tokens serving as its liquidity pool base. Public information indicates the project aims to build a low-cost AI points trading market, offering discounted points for over 400 AI models including Claude Fable and GPT Astra. BlockBeats reminds users that related tokens have low trading volume and high price volatility, urging caution for investments.

Crypto KOL XXAntiWar today showcased their futures trading performance so far this year on the X platform. According to shared screenshots, XXAntiWar’s two accounts have generated a combined profit of over $5 million. The influencer also stated that their largest loss in futures trading this year came from a short position in ZEC. Summarizing their futures trading experience, XXAntiWar emphasized the importance of trading discipline: “Once you have a liquidation price, your trade has a weakness. When you make money, withdraw profits instead of reinvesting them. Put most of your other disposable funds into fixed-term deposits that are not easily accessible, to prevent yourself from overtrading and adding margin. You need to set a limit on your maximum possible loss. For now, the approach that works best for me is to only allocate 3% to 10% of my disposable funds, with a maximum leverage of 3x.”

China's official news agency Xinhua released the full text of the State Council's "Opinions on Developing New Quality Productive Forces". The document proposes to promote the deep integration of the real economy and the digital economy, optimize the layout, construction and operation of national data infrastructure, carry out in-depth digital transformation initiatives for the manufacturing sector, implement projects including the "East Data, West Computing" program, intelligent manufacturing, and industrial internet innovation and development, build a national integrated computing network and a national blockchain network, improve the policy system to advance digital industrialization and industrial digitalization, and foster internationally competitive digital industrial clusters.