SoftBank is seeking to secure an additional $10 billion loan for equity investment in OpenAI.

Insight Beating AI News Flash: Anthropic has launched the Model Hardware Standard (MHS) as a research preview, enabling AI Agents to directly operate real-world machines. Microscopes, liquid handlers, robotic arms, and even laser systems in quantum computers can all connect via a single unified interface. Historically, connecting devices from different vendors required custom interface development, taking weeks or even months. Anthropic says MHS can cut this process to hours, or even minutes. MHS acts as a unified hardware abstraction layer: once devices connect to the MHS driver, they communicate to Agents in a standardized format, detailing their capabilities, current status, and non-negotiable safety constraints. Agents then control devices via MCP, command line, or API, eliminating the need to rewrite connection code for each machine. MHS is not exclusive to Claude—other models and Agents can also integrate with it. Practical tests have been run on real hardware: Carnegie Mellon University integrated an experimental automation system that would normally take weeks to set up, completing the connection in ~8 hours and boosting experimental speed by ~3x. Quantum computing firm QuEra had Claude automatically optimize its laser recovery program, resulting in 695 successes out of 700 blind tests, a 99.3% success rate. AWS, Hugging Face, Raspberry Pi, Universal Robots, and others are also integrating with MHS. However, MHS is not yet a public standard; it is currently available only to select research institutions and vendors via application. Anthropic plans to open-source it after completing additional security testing. At this stage, Claude’s understanding of the real physical world remains limited. During Genentech testing, it failed to independently identify liquid bubbling as a physical fault, requiring researchers to clarify it was not a software issue.

California’s meme coin regulatory bill AB 2409 has passed both the state’s Assembly and Senate, and is now submitted to the Governor’s office awaiting signature. The legislation will ban California public officials and government employees from issuing meme coins, and prohibit digital asset service providers from offering California residents trading services for meme coins issued or co-launched by federal, state, or local public officials starting January 1, 2027. If signed, AB 2409 will become one of the first U.S. state-level regulations targeting politicians issuing meme coins.

According to Bitget market data, Hong Kong stocks closed: the Hang Seng Index rose 0.07% while the Hang Seng Tech Index fell 0.33%. The precious metals sector advanced throughout the trading day, with China Silver Group (00815.HK) and Everest Gold (01815.HK) closing up more than 7%.

Dongcha Beating AI News Flash: The People's Bank of China (PBOC) will issue the "Sci-Tech Innovation Future (Artificial Intelligence)" gold and silver commemorative coin set, consisting of 3 pieces: 1 gold coin and 2 silver coins, all legal tender. The 5-gram gold coin is 18mm square, with designs including a chip, brain-shaped circuit, and 0/1 elements on its reverse side. Each of the two 15-gram silver coins features distinct designs: one with a face, circuit, and spiral dot matrix; the other with a key and AI symbol. The gold coin has a face value of 80 yuan, with a maximum mintage of 10,000 pieces; each silver coin carries a face value of 5 yuan, with a maximum mintage of 20,000 pieces. It should be noted that the face value here is not the selling price. The face value of precious metal commemorative coins mainly serves to indicate their legal tender status, while the actual selling price is determined by factors such as gold/silver raw material costs and production expenses. In March this year, when soliciting coin design drafts, the organizer explicitly allowed designers to use AI tools for assistance, but prohibited submitting AI-generated content directly. The three coins are themed respectively on: basic research in artificial intelligence, "technology for good", and the application of "AI +" across various industries.

According to an official announcement, Binance will launch multiple USD?-margin TradFi perpetual contracts on August 28. The new contract underlying assets include Tempus AI (TEM), Merck (MRK), IonQ (IONQ), MARA Holdings (MARA), and Pinduoduo (PDD). The TEMUSDT perpetual contract will launch at 17:00 Beijing Time on August 28, MRKUSDT at 17:05, IONQUSDT at 17:10, MARAUSDT at 17:15, and PDDUSDT at 17:20. All these contracts are settled in USDT, support up to 20x leverage, operate 24/7, have a minimum notional value of 5 USDT, feature an 8-hour funding rate settlement cycle, and support multi-asset mode.

Beating AI Express News: Tencent’s Hunyuan Hy4 preview not only scales up the model size but also overhauls its underlying architecture. The previous Hy3 generation used traditional full attention, while Hy4 adopts Gated DSA, the gated version of DeepSeek Sparse Attention. When processing extremely long texts, the model no longer recalculates all previous content; instead, it first selects the most relevant parts for focused processing. Hy4 also adds IndexCache to DSA. Originally, each layer of the model had to re-judge which content is most important, but now some layers can directly reuse earlier selected results, cutting down on significant redundant computation. Tencent explicitly stated that this design is inspired by DeepSeek and GLM. Another change is the residual structure: traditional Transformers can be roughly viewed as having only one information backbone, while Hy4 uses iHC to expand this to four backbones, enabling the model to retain more information simultaneously in deep networks. The number of MoE (Mixture of Experts) experts has also increased from 192 in Hy3 to 256 routing experts plus 1 shared expert. Each token selects 8 routing experts and always passes through the shared expert.