SlowMist: FomoPeek Versions 1.1–1.2 Contain Malicious Code, May Result in Leakage of Private Keys and Mnemonics

Beating AI News (from Dongcha): Inco AI has open-sourced its local inference engine Splash, which runs Qwen3.8-27B on an M5 Max MacBook Pro at up to 144 tokens per second. LM Studio has since integrated Splash, and version 0.4.25 is available for direct use. Inco’s earlier DFlash technology has been adopted by SGLang, vLLM, TensorRT-LLM, and llama.cpp. Companies including Meta, NVIDIA, Xiaomi, and Poolside have also integrated DFlash to accelerate their small models. DFlash works by first having a small model predict multiple tokens in parallel, then submitting them to a large model for batch validation, reducing the computation required for token-by-token generation. Splash extends this optimization to its entire local inference engine. It currently only supports Qwen3.8-27B and Qwen3.6-35B-A3B, with each model having dedicated GPU kernels, memory configurations, and DFlash 2 small models. While its model coverage is limited, it delivers aggressive performance optimizations. The 144 tokens per second is the peak demo performance on the M5 Max. On the 48GB M5 Pro used by Inco for cross-device testing, Qwen3.8-27B hits 74 tokens per second for a single short-context run; with 4 concurrent runs, total throughput reaches 170 tokens per second, 3.9 times that of the second-place alternative. This is why Splash is particularly well-suited for agent use cases. When an agent launches multiple subtasks simultaneously, total concurrent throughput is often more critical than the speed of a single conversation. Splash is open-source and not tied exclusively to LM Studio. It currently requires an M3 or newer Mac, macOS 26.4 or later, and at least 36GB of unified memory.

EmberCN reported that Garrett Jin, the agent of the so-called "BTC OG insider whale" entity, was able to absorb tens of millions of dollars in unrealized losses from the ZEC airdrop thanks to its massive early BTC holdings. The entity accumulated 100,784 BTC between May and June 2018 at an average price of roughly $7,242; by the same period last year, this position was worth over $10 billion, and funds for subsequent investments primarily came from BTC sales. Over the past year, the entity has engaged in frequent large-scale crypto asset transactions: starting in August 2025, it sold approximately 89,000 BTC and purchased around 900,000 ETH at an average cost exceeding $3,500 per ETH, with these ETH subsequently transferred to Binance in the first half of 2026. In December 2025, the entity went long ETH on Hyperliquid; in early February 2026, as ETH fell from $3,000 to $1,800, its long position of 213,000 ETH (valued at roughly $670 million) was liquidated, resulting in a loss of about $230 million in margin. During the same period, the entity— which had multiple large fund flows with Trend Research addresses— is believed to have provided partial funding to Trend Research; one of Trend Research’s leveraged long positions of 651,500 ETH was ultimately stopped out, incurring a loss of around $734 million. Additionally, from February to June this year, the entity accumulated a total of approximately 289.6 million "Binance Life" tokens via Binance withdrawals and on-chain purchases, worth roughly $150 million, accounting for 28.9% of the token’s total supply. In March, it suffered extreme slippage when converting 50.43 million aEthUSDT to aEthAAVE, receiving only 327.2 AAVE in exchange and losing about $50.39 million. Since July, the entity has been shorting ZEC on Hyperliquid, with current unrealized losses of around $34 million. Compared to its early $10-billion-level BTC holdings, this loss is just one part of its multiple high-value transactions.

，据 TradingBeats 监测，交易员 Loracle 目前做空闪迪浮亏高达 467 万美元，持仓规模为 2800 万美元，开仓价 1,485.67 美元，爆仓价 2,552.96 美元。与此同时，其地址内持有的 HYPE 空单浮亏扩大至 291 万美元，持仓规模为 700 万美元，开仓价 53.97 美元，爆仓价 136.56 美元。

Beating AI News: Qianwen releases real-time simultaneous interpretation model Qwen3.8-LiveTranslate. The model cuts the average lag time (LAAL, a metric measuring simultaneous interpretation average delay) from 2.8 seconds of its predecessor to 2.3 seconds, and adds new capabilities including real-time speaker diarization, synchronized source and translation output, and long-context disambiguation. It can now distinguish different speakers, align translations with specific utterances, and ensure translated speech more stably retains each speaker’s voice characteristics. Source and translation texts are also returned in a synchronized stream. The model further leverages previous context to identify names, terms, and references, reducing ambiguity stemming from only processing the current sentence. API pricing remains largely unchanged versus the previous generation. Per Alibaba Cloud’s price list, the prices for audio input, image input, text output, and audio output of Qwen3.8-LiveTranslate in the Beijing region are identical to those of Qwen3.5-LiveTranslate; the four prices for the Singapore region have even decreased slightly.

According to monitoring by AI Auntie, a certain entity has bought an additional 2,086 ETH at an average price of $2,599 over the past 15 hours, totaling around $5.42 million. Since yesterday, the entity has been adding to its ETH position by selling UBTC, having purchased a cumulative 9,058.19 ETH for approximately $22.57 million at an average cost of roughly $2,492.62, with an unrealized profit of about $1.22 million at present.

Per HTX market data, the wallet address bc1qsz associated with BIT (formerly Matrixport) has deposited an additional 1,000 BTC into Binance, worth roughly $81.06 million. On September 15, BIT (formerly Matrixport) deposited 1,400 BTC into Binance, valued at approximately $107.8 million, while withdrawing 10,000 ETH from the exchange, worth around $24.67 million.