Vitalik: The zkEVM speed race is completely over, and the primary task should fully shift to security

After briefly issuing a rumor-clarification statement titled "Cheung Kong responds to Li Ka-shing’s death: No such thing", mainstream Chinese financial media outlets including The Paper and Jinshi Data have collectively deleted the related news report. As of now, Caixin News Agency still retains the report, but its source is also The Paper. Some media that reposted the content have since re-marked the source as The Paper.

Beating AI News Flash: StepFun has open-sourced the Step 5 Preview model weights, with FP8 and BF16 versions now available on Hugging Face and ModelScope, while supporting inference code has also been released simultaneously. The Step 5 Preview model has a total of 600 billion parameters, with only approximately 27 billion activated per generation, and supports a 1-million-token context window. A newly released technical blog reveals the model adopts a 92-layer "narrow and deep" architecture, allowing information to be processed through more layers to support multi-step reasoning in complex tasks. It also introduces Sparse Grouped Query Attention (Sparse GQA), merging indices of 8 adjacent tokens to reduce indexing and Top-k filtering overhead to 1/8 of that of per-token indexing schemes. During the reinforcement learning phase, the team incorporated context compression into training, enabling the agent to organize progress when tasks are too long before resuming execution. The model uses the custom StepFun Model License 1.0, which permits general commercial use, but hardware manufacturers such as mobile phone and automotive companies must sign a separate license agreement if integrating the model’s capabilities into commercial products. Enterprises operating MaaS (Model as a Service) businesses with total revenue exceeding $20 million over 12 consecutive months (including their own and affiliated entities) also need to obtain separate permission for commercial use.

According to TradingBeats monitoring, Hyperliquid’s Shanghai-listed Korean stock mapping contract SKHX fell approximately 10.46% since 00:00 today, currently trading at $938.71, while its US-listed ADR mapping contract SKHY dropped around 7.80% over the same period, with a 2.66 percentage point gap in declines. Based on the conversion ratio, SKHY’s implied premium relative to SKHX has expanded from roughly 42.89% at 00:00 today to 47.14%, a rise of about 4.25 percentage points, with the premium itself increasing by around 9.9%. Over the same period, SKHX fell 10.46% while SKHY dropped 7.80%, and the latter’s relative resilience further widened the price gap between the two contracts. Against the backdrop of the price divergence between the two regions, a whale trader who shorted the Korean stock and went long on the US ADR has reaped substantial profits. The account currently holds: - SKHX short position: $17.59 million, average price of $1,293.84, with an unrealized profit of approximately $6.6549 million; - SKHY long position: $6.59 million, average price of $175.04, with an unrealized loss of around $1.7636 million. After netting the two legs, the unrealized net profit is roughly $4.8913 million. This account is not a fully hedged spread trade. Converted based on the number of underlying shares, its ADR long position only covers about 25.5% of the Korean stock short position, leaving a large net short exposure overall. A large portion of the current profit still comes from SKHX’s sharp decline. It is reported that the whale also added approximately $356,100 to its SKHX short position today, while reducing its SKHY long position by around $1.3973 million, further increasing the directional short bias of the portfolio. After including another DRAM short position with an unrealized profit of about $315,100, the net unrealized profit of the account’s three positions totals approximately $5.2063 million.

According to TradingBeats monitoring, SKHX on Hyperliquid has dropped nearly 15% in the past 24 hours, currently trading at $939.46, down roughly 12.88% from its level when last night’s CPI data was released. Amid the sharp plunge, large long positions have been liquidated one after another, incurring losses. Since last night to this morning, three verified whales have closed a total of approximately $3.881 million worth of SKHX long positions, with combined losses of around $887,000 via both stop-loss orders and forced liquidations. One whale liquidated roughly $1.493 million in long positions, taking a loss of about $347,000 (address: 0xcb13); another long trader reduced positions via voluntary loss, but their remaining positions were eventually forced liquidated, with total losses of around $261,000 in this round (address: 0x5fc2). The third whale’s approximately $1.008 million in long positions was also liquidated in batches, adding another loss of roughly $279,000 (address: 0x3b5b). All three accounts currently hold no SKHX positions. However, after the plunge pushed funding rates into negative territory, bargain-hunting funds have since flowed back. SKHX’s hourly funding rate reversed rapidly: from -0.08399% at 08:00 to +0.03080% at 09:00, and further rising to +0.03550% at 10:00, flipping the long-short pricing dynamic for the contract. Long liquidation risks remain unresolved. Liquidation distribution data shows that around $5.09 million in potential long liquidation exposure is clustered near the $888 level for SKHX, roughly 5.5% below the current price. If the price continues to drop, this area will be a key watch point for the next round of leveraged long positions being forced out.

Rumors are circulating in the market that 98-year-old Li Ka-shing has passed away. An insider from Cheung Kong Holdings (01113.HK) responded, saying: "This is completely untrue."

Beating AI News Flash: Ant Group’s Ling large model has announced the open-source release of Ling-3.1-flash. Positioned as "truly usable", the model features upgraded capabilities for real-world tasks, enabling faster conversion of ambiguous business requirements into actionable results. The open-source model’s weights are available for download on Hugging Face and ModelScope. Ling-3.1-flash inherits the mixed linear architecture of Ling-3.0-flash, boasting approximately 560B total parameters, with around 25B parameters activated per token, and supporting up to a 1M-token context window. The model delivers high-quality performance across complex workflows including office work, healthcare, software engineering, and terminal operations. When handling scattered data, it can assist in organizing reports and presentation materials; for fragmented symptoms, it helps sort out medical histories and consultation issues; for legacy code, it can understand existing implementations, write new features, and revise them based on test outcomes. In the office task evaluation GDPVal-AA V2.1, Ling-3.1-flash scored 1673 ELO points. In the HealthBench Professional medical scenario benchmark, it achieved a score of 65.35. It also demonstrated strong proficiency in complex engineering tasks in the software engineering benchmark FrontierSWE and terminal operation benchmark Terminal-Bench 4.0.