10 #Bitcoin ETFs inflows 2,586 $BTC(+$247.92M) and 9 #Ethereum ETFs outflows 20,884 $ETH(-$70.69M) yesterday.

HashKey Holdings Limited today announced its unaudited consolidated interim results for the six months ended June 30, 2026 (the "Report Period"). During the period, the group achieved total consolidated revenue of HK$343 million, up 20.6% year-over-year; gross profit reached HK$208 million, rising 12.5% YoY; gross margin improved sequentially to 60.6% from the second half of 2025. As the group’s core growth engine, transaction facilitation service revenue hit HK$268 million, surging 38.6% YoY. Total platform trading volume rose against the trend to HK$282.2 billion, up 31.8% YoY. Institutional client trading volume continued to grow, jumping 58.8% YoY to HK$231.5 billion, accounting for 82.0% of total trading volume, reflecting institutional investors’ high confidence in compliant platforms. Beyond the trading segment, the group’s on-chain services and asset management businesses also made phased progress. During the period, HashKey launched Hong Kong’s first real estate RWA and first regulated silver RWA token, further consolidating its industry-leading position in on-chain financial innovation. Total on-chain RWA value reached HK$2.68 billion, up 167.8% YoY. Additionally, the group achieved notable growth in its asset management and investment segments. Its assets under management (AUM) stood at HK$5.94 billion, with revenue of HK$38.84 million. In the first half of this year, HashKey strategically invested in Vietnam’s crypto asset exchange Shengxingwang (CAEX), while a fund under HashKey Capital Investment led SignalPlus’ Series B+ round with a US$40 million investment.

Beating AI Express: IBM has open-sourced its new-generation Granite 4.2 language model, available in three variants: 3B, 8B, and 30B. All three models feature native reasoning, capable of switching between full reasoning, low-intensity reasoning, and no reasoning, as well as direct tool calling. The 8B and 30B versions additionally come with dedicated agent reinforcement learning, enabling them to execute tasks in real code repositories, terminals, and web search environments. Training rewards are directly tied to task outcomes, such as whether code passes tests or terminal tasks are completed. All three models are released under the Apache 2.0 license. Independent evaluations show the small models performed surprisingly well: Artificial Analysis awarded the 3B version 14 points, ranking it 2nd among 46 open-weight models of the same parameter size; the 8B version scored 20 points, significantly higher than the median of 9 for its peer group. In IBM’s internal tests, the 30B version achieved 57% on SWE-Bench Verified and 89.17% on AIME25. There is an architectural shift: Granite 4.0 featured a Mamba-2/Transformer hybrid architecture, which IBM claimed reduced memory requirements by over 70% in long-context and multi-concurrency scenarios; Granite 4.2 has switched back to a pure dense Transformer architecture. Compared to the previous generation’s focus on inference efficiency, this generation clearly prioritizes inference and agent capabilities.

According to TradingBeats (formerly Hyperinsight) monitoring, the arbitrage whale with the address starting with 0xf17, which had previously been continuously reducing HYPE’s spot-futures hedge positions, has now shifted to directional long positions, pushing its net long exposure on HYPE to around 222,000 coins. In the early hours, the whale first bought 39,700 HYPE spot coins in a concentrated 7-minute window, with a trading volume of roughly $3.251 million at an average price of $81.98. It then opened 136,000 long positions in HYPE perpetual contracts, totaling $11.026 million at an entry average of $81.05. Combined, the two purchases amounted to approximately $14.277 million. As of press time, the whale’s holdings stand at: 86,000 HYPE spot coins worth ~$7.008 million; 10x fully margined long positions of 136,000 HYPE perpetual contracts, valued at ~$11.087 million, with an entry average of $81.05, unrealized profit of ~$61,000, and liquidation price of $38.64. The total value of its HYPE spot and futures positions is ~$18.095 million. Calculated at current prices, the new spot and perpetual longs added yesterday have generated ~$42,000 in temporary unrealized profit. Since August 18, this whale has repeatedly traded HYPE in band long positions, with cumulative profits from closed related transactions reaching ~$320,000. It currently has 100 spot sell orders placed in the $83.5–$110 range, planning to sell 50,000 HYPE coins for a nominal ~$4.837 million, at a weighted average sell price of $96.75, covering ~58.1% of its spot holdings. HYPE is now trading at $81.50, down ~0.9% in 24 hours. Its perpetual futures open interest is ~$2.086 billion, with a trading volume of ~$752 million, spot volume of ~$118 million, and current hourly funding rate of +0.00125%. Prior news: HYPE’s arbitrage space narrowed, funding rate dropped, leading the whale to withdraw $11 million in hedge positions.

According to monitoring by YuEmber, as Changxin Technology (CXMT) saw its A-share price surge by 5% in today’s trading, the largest short seller who opened short contracts one day before CXMT’s A-share listing has racked up total losses of $10.82 million. Of this sum, $6.3 million stemmed from price fluctuations, while the remaining $4.52 million was incurred as funding fees.

According to an official announcement, due to abnormal situations including price deviations of TAC (TAC Protocol) in the spot markets of multiple centralized exchanges (CEXs), Binance Futures will apply the latest Price Protection (LPP) mechanism to the TACUSDT U-margined perpetual contract at 19:00 (GMT+8) on August 27, 2026. The announcement noted that during the switch to the LPP mechanism, to minimize the impact of sudden price fluctuations and better ensure price stability throughout the transition, the LPP mechanism will be gradually adopted via a marked price smoothing mode. The LPP cycle for the contract will end when TAC’s prices across multiple exchanges converge and a stable index price is available in the spot market (as determined by Binance). A separate announcement will be released once the LPP cycle for the TACUSDT U-margined perpetual contract concludes and it is ready to transition from LPP to the standard marked price calculation (marked price = Median (Price 1, Price 2, Contract Price)). The transition duration will depend on price volatility and the stability of the index price.

Insight Beating AI Flash News: Instinct, a personal AI assistant valued at $2.5 billion, has 23-year-old founder Noah Shinn. The Wall Street Journal reports Shinn dropped out of Northeastern University in 2023, after previously conducting machine learning and programming language research at both Northeastern and MIT. His most notable work is Reflexion, for which he served as first author of a 2023 NeurIPS paper proposing that agents summarize errors after failure, form memories, and apply these to subsequent tasks. The paper achieved a 91% pass@1 on HumanEval, outperforming the 80% result reported for GPT-4 in the paper. He later co-developed τ-bench, an agent evaluation framework that tests whether models can simultaneously communicate with users, call tools, adhere to business rules, and maintain stability across multiple runs. Shinn then joined enterprise AI agent firm Sierra as one of its earliest employees, working as a research scientist. This career path nearly directly leads to today’s Instinct: agent reflection, tool calling, real-task evaluation, enterprise agents, and finally launching his own personal agent. After Instinct’s official announcement, figures including Conviction founder Sarah Guo, Altimeter founder Brad Gerstner, and former Yandex Head of Search, AI, and Cloud Andrey Styskin extended congratulations in the comments section.