The day before yesterday, a whale that shorted Bitcoin with 40x leverage was forced to reduce its position, incurring an unrealized loss of $1.54 million.

UBS CEO Sergio Ermotti said today that despite mounting geopolitical and economic risks, investors have grown more complacent over the past few years, and the current market should have seen far greater volatility. Ermotti noted that energy and shipping risks stemming from conflicts in Iran and Ukraine, persistent global supply chain pressures, high inflation, and elevated borrowing costs remain in place. While large-scale investments in areas like artificial intelligence (AI) and data centers have supported economic growth, they are not enough to eliminate underlying risks. On investment strategies, he said UBS clients have recently favored cross-sector and cross-region diversified allocations over bets on a single market direction, while continuing to hold AI and tech assets. He stressed this does not mean funds are fleeing the U.S. or the dollar en masse, as the dollar remains the world’s primary "reference currency". Regarding interest rates, Ermotti expects the European Central Bank (ECB), U.S. Federal Reserve, and Bank of Japan (BOJ) may all raise rates in the coming months, warning markets not to anticipate a quick return to low borrowing costs. He added that inflationary pressures have not eased significantly, so it is reasonable to expect interest rates to stay elevated for some time.

Insight: Beating AI Flash News – DeepSeek V4.1 Flash has nearly overhauled its entire architecture this time, aiming to simultaneously enhance performance, boost speed, and cut costs. Stronger: The model features 552 billion backbone parameters, plus an external Engram conditional memory module with 196 billion parameters. Pre-training utilized 45T multi-modal tokens, while post-training integrated a large volume of real agent tasks, tool environments, and failure cases. DeepSWE v1.1 scores 74.2%, surpassing Claude Opus 5 and GPT-5.6 Sol. Faster: The new CED architecture splits the 40-layer model into two halves. When reading prompts, only 8 billion parameters are activated per token, and just 16 billion during generation. Coupled with CSA2 cross-layer reuse and DSpark speculative decoding, the context window has been expanded from 4K to 1M (256x longer), with decoding computation per token increasing by only roughly a quarter. Cheaper: DeepSeek has further optimized KV cache to the maximum. The main cache is switched to FP4, plus cross-layer reuse, leaving global KV for each token at just 890 bytes, about 1/4 of V4 Flash; cache stored long-term on SSD or memory is further reduced to approximately 1/8. V4.1 Flash does not equate "stronger" simply to "more computation". While model scale continues to grow, only a small portion of parameters are adjusted each time; the context window is extended, yet the cache is compressed even smaller. Performance has improved, with speed and costs remaining uncompromised.

According to Arkham, Bitwise’s BSOL ETF has posted a net inflow of $107.4 million over the past 20 trading days, with only two trading days recording capital outflows.

August 2026 witnessed a notable recovery in cryptocurrency market trading activity, with core data as follows: Spot-led growth: Spot trading volume on major centralized exchanges (CEXs) rose 18.64% month-over-month, while perpetual contract volume increased 10.89%. The significantly higher spot growth indicates the market rebound is primarily driven by spot buying demand. Traffic-trading divergence: Major CEXs saw a 0.26% slight drop in website visits in August, and app downloads fell 5.61%. Against the backdrop of slowing new user traffic, trading volume rose against the trend, signaling this round of growth mainly comes from increased trading frequency and conversion efficiency among existing users. Perp DEXs also rebounded: Major perpetual decentralized exchanges (Perp DEXs) recorded an 8.98% month-over-month rise in August trading volume, with overall web traffic remaining stable. CEX rankings & performance: Volume leaderboard: Binance remains the industry leader with $233 billion in spot volume and $1.606 trillion in contract volume. OKX (spot: $46 billion / contracts: $634 billion) and Bybit (spot: $46 billion / contracts: $365 billion) rank second and third respectively. Rank changes: Coinbase’s spot ranking climbed 2 spots to 7th; Kraken rose 1 spot to 9th. User traffic leaderboard: Binance (2.532 million downloads, 34.19 million web visits) and OKX (550,000 downloads, 25.62 million web visits) retain the top two positions. Perp DEX highlights: Hyperliquid leads the decentralized perpetual contract market with $207 billion in contract trading volume. EdgeX’s trading volume surged 70.60% month-over-month; ApeX, Lighter, and Aster all posted notable growth of 17% to 22%.

Beating AI News Flash, citing The Wall Street Journal (WSJ): South Korea’s government is set to announce a major energy investment project in the U.S. to support American artificial intelligence (AI) development, a long-awaited milestone from the trade deal Washington and Seoul struck last year. According to sources familiar with the matter, the deal has a potential value of over $100 billion. The plan calls for South Korea to fund the construction of up to eight nuclear power plants and one natural gas project. Last October, former U.S. President Donald Trump and South Korean leader Lee Jae-myung reached an agreement: the U.S. would lower most tariffs from 25% to 15%, including those on South Korea’s auto sector. In exchange, South Korea pledged $350 billion in U.S. investments plus an additional $100 billion in purchases of American energy. No projects had been implemented prior to this, drawing dissatisfaction from U.S. officials including Trump. Trump had threatened in January to raise tariffs on South Korea again. The two sides are now close to finalizing two agreements, which could be announced as early as next week. Negotiations remain ongoing, and investment details as well as the announcement timeline may change.

Beating AI Express News: DeepSeek Harness v0.1.5 has been released. The V4.1 Flash model underwent dedicated training and optimization specifically for Harness during development, covering three modes: Standard, PTC, and Minimalist. Harness is now directly integrated into the model training process, rather than merely serving as an Agent wrapper added post-model release. The new version also introduces experimental Agent Teams, enabling bidirectional communication between parent and child Agents. The main Agent can assign models and inference intensity to different child Agents; multiple Agents can also share tasks, send messages to each other, and have their outputs aggregated by the main Agent. V4.1 Flash features a notable long-task optimization: when an Agent modifies system prompts mid-run, it can retain most existing KV Cache, eliminating the need to recalculate large subsequent context segments due to minor rule adjustments earlier in the task. The web version now adds file upload and preview functionality, supporting formats including PDF, images, and code. Plugins can integrate their own UIs into the left/right sidebars and content areas, while optimizations have been implemented for long sessions in terms of loading, recovery, and memory usage.