10 #Bitcoin ETFs outflows 4,964 $BTC(-$458.3M) and 9 #Ethereum ETFs outflows 4,701 $ETH(-$15.21M) yesterday.

According to HTX market data, likely driven by news that Upbit will list CT on its KRW, BTC, and USDT trading pairs, CT surged over 48% in a short period, briefly breaking above $0.44, and has since pulled back to $0.39, with its fully diluted valuation (FDV) rising to $300 million.

Morgan Stanley raises Meta Platforms (META.O) target price from $775 to $820. (Jinshi)

According to HTX market data, Bitcoin briefly broke through $83,000, with a 24-hour gain of 1.54%.

Upbit will list CT on its KRW, BTC, and USDT markets.

According to Coinglass data, UNITREE saw $1.9284 million in total liquidations across the network over the past 12 hours, with short positions accounting for $1.9049 million and long positions only $23,500. Unitree Technology’s stock rose 16.15% today, pushing its market capitalization to 200.58 billion yuan. Meanwhile, UNITREE perpetual contracts on Hyperliquid have recorded five consecutive negative funding rate settlements between 10:00 and 14:00 today, with a cumulative rate of approximately -0.47087%. Hyperliquid’s current funding rate is around -0.13732% per hour, with the estimated annualized funding rate rising to 1,240%. On Binance, UNITREE contracts currently carry a rate of approximately -0.777501% per 4 hours, with the next settlement scheduled for 16:00 Beijing Time, roughly 38% higher than Hyperliquid’s rate. If prices and funding rates remain at current levels, the 24-hour funding cost for a $10,000 short position would be around $338 on Hyperliquid and $467 on Binance.

Insight: Beating AI Flash News, Google Research has open-sourced Harness2, enabling AI agents to automatically modify system prompts, memory, skills, tool rules, and even scripts and plugins without retraining the underlying model. The project is licensed under Apache 2.0 and supports OpenCode and Codex. Its core mechanism lets agents learn through trial and error: when facing a new task, the system first proposes multiple improvement plans, has different agent versions execute them separately, compares their performance, and finally combines the best elements to form a more suitable solution. Even without a standard answer, it can identify improvement directions based on actual execution processes. After a task is completed, the system retains records of which modifications worked and why, so it can reference this data for similar tasks in the future instead of starting from scratch each time. Google tested Harness2 using GLM-5 and Gemini 3.5 Flash. On WorkBuddy-Bench, GLM-5 paired with OpenCode saw its average score jump from 64.2 to 75.7 after just one round of optimization, outperforming unoptimized Gemini 3.5 Flash’s 72.4 points. In legal task tests, GLM-5’s pass rate for some scoring criteria also rose from 59.6% to 71.4%. However, the experiment defaulted to Claude Opus 4.8 for proposing and implementing improvements, with GLM-5 and Gemini primarily responsible for executing tasks—meaning the evolution was not achieved entirely by these models on their own. Chinese developer Joseph Zhou also demonstrated an experiment integrating Harness2 with Jev 1.14, noting consistent improvements in performance on visual tasks.