In the past hour, total liquidations across the network reached $16.86 million, with $1.83 million from long positions, with $15.03 million from short positions. BTC liquidation $2.44 million, ETH liquidation $4.49 million.

According to data from Polymarket, as reported by Odaily Planet Daily, account betoor619 (address: 0x70088c990ffae782c699b9250f5aa6cbe4e3c666) bet on Spain to win in the World Cup group stage match between Spain and Cape Verde. The match ended in a 0-0 draw, resulting in a loss of $999,000 for the user.

Odaily Planet Daily reports that Iran's Deputy Foreign Minister stated that the next phase of negotiations will cover nuclear issues including uranium enrichment, nuclear stockpiles, and Iran's nuclear needs. (Jinshi)

According to a post on the X platform by Citrini analyst Jukan, as reported by Odaily, South Korean media outlets indicate that SK Hynix is expected to announce a massive shareholder return plan worth approximately $66.4 billion after its ADR listing. To mitigate dilution concerns and potential opposition from existing shareholders, the company has reportedly reduced the ADR offering size from 2.4% of its outstanding shares to just over 2%. SK Hynix expects to raise approximately 40 trillion won, or about $26.5 billion, through the new share issuance in the ADR offering.

According to CoinWorld, Redis founder Salvatore Sanfilippo refuted claims that large-scale Chinese models rely on distilling American models. He pointed out that ordinary APIs can only return textual answers and cannot replicate the probability distribution and internal state of the model when generating the answer. The logical reasoning and thinking abilities of large-scale models are hidden in complex neural networks; external users can only obtain the final textual answer, but not the complete thought process and probability calculation process. He emphasized that the rise of large-scale Chinese models like DeepSeek relies on solid accumulation of underlying engineering such as data pre-training and reinforcement learning, rather than taking shortcuts. The academic community divides large-scale model distillation into "soft distillation" and "hard distillation," with current debate mainly focusing on "hard distillation." Large companies are guarding against API abuse, attempting to protect their first-mover advantage.