A whale address dormant for four years has woken up and transferred 4,500 BTC.

Beating AI Express News: DeepSeek’s official desktop client for Harness has leaked early. DeepSeek has not made an official announcement, nor has it launched a public download portal on its website, but macOS and Windows installers have appeared on its official servers and are confirmed to install and work properly after testing. Built on Electron, the desktop client packages Harness’s original web version into a standalone app. Users can log in with a DeepSeek account or input an API key, select a local folder as their workspace, and do not need to manually start a web service. The official repository has recently been rapidly adding desktop client features. The latest dsh v0.1.7-rc.2 adds onboarding guidance, shortcut key management, and background task functionality, and fixes issues including Windows installer startup and macOS window problems. Currently, only macOS Apple Silicon and Windows x64 official installers are available. The Mac version is signed with Hangzhou DeepSeek’s Developer ID and has passed Apple notarization.

According to an official announcement, Binance Wallet has launched a no-lock interest-earning feature. Users who hold eligible stablecoins in their Binance Wallet and activate the feature can earn rewards without staking, locking funds, or interacting with smart contracts. Currently, three assets are supported: U, USDe, and USDS. U is supported on the BSC network with an estimated annualized reward of 1.5%; USDe is available on Ethereum with an estimated 4.75% annualized reward; USDS, also on Ethereum, offers an estimated 3.6% annualized reward. Binance noted that rewards begin accruing 24 hours after activation of the feature and holding eligible assets, with rewards accumulated daily and disbursed weekly. The rewards are funded by partner promotion campaign funds and do not stem from returns generated by the assets themselves. Furthermore, Binance Wallet will soon launch a U-based leaderboard campaign with a total prize pool of $150,000, to be distributed in U tokens.

Xie Jiayin, head of Bitget’s Chinese-speaking region, stated in a post that the exchange has commissioned a third-party security team to conduct an independent forensics investigation into the incident. Multiple teams are collaborating to verify remediation measures, confirm the attack vector, and assist with law enforcement cooperation. Once the matter is concluded, the investigation results will be shared in a timely manner. Withdrawals will be reopened immediately after all potential risks are eliminated.

Bitget announced on social media that its Protection Fund holds 5,500 BTC, worth approximately $464 million at current prices. All fund wallet addresses are public, and anyone can verify them on-chain at any time. The fund is designed to protect user assets from platform-level threats such as cybersecurity incidents. Losses from this hot wallet incident will be covered by the Protection Fund after assessment. The exchange will replenish the fund, with specific details and coverage terms to be announced separately.

Beating AI Express News: NeoHorse-Jev-4B, an open-source decision-making model from Jiyuan Ludong (founded by Wang Yunhe, former head of Huawei’s Pangu large model), is built on the company’s NeoHorse-1-4B. Unlike standard large models that generate answers token-by-token, this model selects from multiple candidates, judges correctness, or assigns scores, primarily for agent tool selection, routing, and next-step action decisions. The developer aggregated results from six text benchmarks including JevBench, Kev, and OpenJev. NeoHorse-Jev-4B scored an equal-weight average of 77.70, outperforming the 9B-parameter Open-Jev (75.67) and Kev-4B (74.25). Note that the composite score was compiled by Jiyuan Ludong itself; currently only four open-source models have complete results across all six benchmarks, and NeoHorse-Jev only took top scores in two of them. Built on NeoHorse-1-4B, it uses prefill-only inference to directly calculate candidate answer probabilities, eliminating the token-by-token generation step common in standard LLMs. In tests with 1,158 text requests, the server recorded an average decision latency of 74.3ms at 32 concurrent connections, and a throughput of 347.05 requests per second at 256 concurrent connections. The model supports text and single-image inputs. Its weights, inference code, and deployment examples are all open-source, runnable via vLLM, SGLang, or local Runtime, and licensed under Apache 2.0.

According to on-chain data analyst Yu Jin Monitoring, a whale or institutional address transferred 541,000 HYPE tokens (valued at approximately $49.54 million) to Kraken 15 minutes ago. The tokens were withdrawn from Coinbase Prime half a month prior at an average price of $73.9 per unit. Transferred to Kraken at the current $91.5 per token price, the move is estimated to have generated a profit of $9.52 million.