Hong Kong stocks closed lower, with the two leading large language model giants slumping. Zhipu plunged 12.4%, and MINIMAX-W fell 3.89%.

Per TradingBeats monitoring, an address that has held and remained inactive since purchasing and withdrawing ETH four years ago first transferred 0.01 ETH to Coinhako’s deposit address as a test, then sent 8,249.86 ETH at 14:49, valued at approximately $22.64 million, with its balance nearly emptied after the transfer. It is reported that this entity had three associated addresses that collectively withdrew 4,615.15 ETH from Binance back in 2022, and spent about 9.0009 million USDT via four on-chain transactions to acquire 3,635.02 ETH. These funds were aggregated to the address 0xcbe3 in August 2023. Calculated based on the price at each withdrawal time, the reference cost of this batch of ETH is roughly $2,181.8 per ETH, with an unrealized profit of around $4.6359 million, a return of approximately 25.76%, leading to suspicion that the address has completed selling. Address: 0xCbE36050a50E3d3e6077494f4d9Ad81Ec3A6E7d4

Beating AI Insight News Brief: Tesla CEO Elon Musk told CCTV Finance in an exclusive interview that China’s overall large AI models are “excellent”. Notably, even with limited computing power, the performance per unit of compute of Chinese models is already near the global top level. Musk assessed that China will resolve the computing power bottleneck faster than most people expect. He roughly estimated that China can fill its computing power gap in the next 2 to 3 years by leveraging lithography technology and chip manufacturing capabilities. Per his original remarks, he emphasized performance per unit of compute – how powerful a model can be with the same amount of computing power.

Aave v4’s deposit volume has exceeded $1 billion, doubling over the past month, with its current lending volume standing at $300 million.

Beating AI Express: Y Combinator S26 startup Context.dev has launched Scrape Anything, consolidating previously fragmented web scraping APIs into a single interface. Developers only need to provide a URL for a webpage or online file, select their desired output format, and receive results including Markdown, HTML, screenshots, page images, raw files, or structured fields in one request. In addition to standard web pages, the tool currently supports around 40 file formats, such as PDF, DOCX, PPT, and XLSX. JavaScript rendering, proxy switching, anti-crawling handling, and retries are all managed by Context.dev, eliminating the need for developers to maintain their own browser and proxy infrastructure. While these capabilities already existed prior to this launch, the update primarily unifies disparate inputs and outputs under a single /v1/web/scrape endpoint. A standard scrape consumes 1 credit; requests for multiple output formats share this single base charge. Using browser-based operations costs 2 credits, and PDF OCR is billed separately based on the number of pages processed.

According to on-chain analyst Ai Yi (handle @ai_9684xtpa), a whale holding a $4.5M concentrated position in Robinhood-themed meme coins and DeFi blue chips has liquidated its PONS token holdings. The $3.67M sell pressure pushed the token’s price down 4%, from $0.6968 to $0.6688. Between 14:21 and 15:17 today, the whale sold 5.338 million PONS at an average price of $0.6779, totaling $3.67M. PONS was previously the address’s largest on-chain asset, with an entry price of around $0.7154, resulting in an approximate $200,000 loss from this sale.

OECD projects that the U.S. Federal Reserve and European Central Bank will raise interest rates once more this year, then hold rates steady through 2027. It forecasts the Bank of Japan will lift its key rate to 2% by the end of 2027. The Bank of England is expected to keep rates unchanged for most of 2027, with its next policy move likely a rate cut.