Hong Kong stocks closed, with 智谱 down 11.01% and MINIMAX-W down 12.29%.

According to Digital Asset, South Korean crypto influencer "Coin King Jonbur Kim" has been acquitted by the Seoul Southern District Court over his alleged involvement in a crypto scam worth 340 billion won (approximately $253 million). The court announced Thursday that the prosecution’s evidence was insufficient to prove his guilt. The influencer’s real surname is Park; he was previously accused of involvement in issuing and launching fraudulent tokens, spreading false information, and manipulating coin prices in cases involving projects such as Podo Coin and Artube. The court’s ruling holds that existing evidence is insufficient to support the relevant criminal charges.

US September CPI rose 0.6% month-over-month and 3.7% year-over-year, both exceeding expectations, yet Treasuries saw no sell-off. On Thursday, the US 10-year Treasury yield fell as much as 8.4 basis points to 5.105%, hitting a three-week low. Ongoing declines in AI tech stocks and rising associated credit risks drove capital back into Treasuries, outweighing rate-hike fears sparked by hotter inflation. Market concerns are spreading from overvalued AI stocks to the debt financing model for data center construction. Data center operator CoreWeave’s debt financing size is nearly $48 billion, with some of its bond yields rising to around 15% to 16%, sparking concerns over the AI sector’s debt-servicing capacity and private credit risks. BlackRock has downgraded its rating on US equities to "neutral", while upgrading its outlook on Treasuries and showing a preference for short-term Treasuries. Analysts note that recent trends show Treasuries are regaining their safe-haven status during stock market declines, suggesting the traditional stock-bond hedge relationship may be recovering. However, US inflation remains above the Fed’s 2% target, and fiscal deficit and Treasury supply pressures have not abated, so whether Treasuries can sustain their safe-haven function remains to be seen.

Apple Executive Chairman Tim Cook revealed during an interview in Shanghai that Apple Intelligence has been approved for launch in China. A test version will roll out later in October, with initial features including real-time translation, writing assistance, and image editing; Siri AI is not part of the initial offering.

Beating AI News Flash: OpenCode has announced limited-time free access to DeepSeek V4.1 Flash. Users can now select "DeepSeek V4.1 Flash Free" from the model list, with no charges for input and output tokens. However, the free version only offers a 200,000-token context window, an 80% reduction from the paid version’s 1 million tokens. When handling large code repositories or long-running coding agents, the free version is more prone to hitting the context limit. DeepSeek V4.1 Flash was already integrated into OpenCode as early as September 10. On October 12, OpenCode had just raised the V4.1 Flash usage quota in its $10 monthly Go plan to a value of $80, and has now added a separate free tier. OpenCode’s official documentation confirms this free access is time-limited, though no end date has been disclosed. The platform also promises the free tier follows a zero-data-retention policy, and will not use user content for model training.

According to monitoring by Onchain Lens, BitMEX co-founder Arthur Hayes purchased approximately $888,000 worth of crypto assets through FalconX and Flowdesk, including 546,600 AERO tokens valued at around $459,800, 120,000 PENDLE tokens worth roughly $231,600, and 3 million ZAMA tokens valued at approximately $196,700.

Beating AI Express (Dongcha) News: Anthropic has added Idle Compaction to Claude Code, a feature designed specifically to reduce token consumption after long session interruptions. Previously, when users step away from their computers for a period and return to their conversation after prompt caching expires, the model has to reprocess the large volume of prior content. Now, Claude Code automatically compresses long conversations into summaries before the cache expires. When users come back, the model only needs to process the summary instead of the full conversation history, cutting the cost of subsequent API calls. This feature is currently available for the terminal version, with default settings requiring sessions to use 1 hour of prompt caching, a context window of at least 200,000 tokens, and the computer to remain awake. Compaction triggers after approximately 54 minutes of continuous inactivity. However, the cost savings come with tradeoffs: after compaction, Claude can only operate based on the summary, potentially forgetting specific prior instructions, technical details, and decision-making processes. As early as early October, developers complained on GitHub that they were running around 20 long sessions simultaneously, and 7 of them were automatically compacted overnight, forcing them to re-add key context. Anthropic later added an off switch: users can set `"idleCompaction": false` in their `settings.json` file, and also adjust the default 200,000-token trigger threshold.