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

Beating AI News Flash: DeepSeek announced that its API now supports web search and crawling, with billing based on request volume, priced at 5 yuan per 1,000 successful requests. Currently, when users utilize DeepSeek API in DeepSeek Harness, Claude Code, or Codex, they access the Search/Fetch interface via server-side tool calls. These Search/Fetch calls are currently free of charge, and will be billed per use starting from 00:00 Beijing Time on November 1, 2026.

Analysis firm Sentora reported that 89% of the roughly $1.63 billion in liquidations on Ethereum lending platforms over the past year were settled via atomic paths. Atomic liquidations depend on decentralized exchange (DEX) liquidity. Real World Asset (RWA) collateral, which features thin liquidity pools and whitelist transfer restrictions, will push more liquidations to shift to off-chain market makers, request for quote (RFQ), and solver paths. Atomic paths refer to liquidations that complete collateral liquidation and debt repayment within a single on-chain transaction, with common methods including flash loans, self-funded advances, or direct contract calls. By contrast, RWAs are often hard to offload immediately on DEXs, so liquidations will more frequently be handled by compliant off-chain market makers, who acquire the collateral via inquiry transactions and solvers before closing out positions.

Last week, US tech stocks saw a sharp pullback, with the Nasdaq down over 4% for the week. While digesting stronger-than-expected inflation data, markets are re-examining the "unbreakable economy" narrative. The Federal Reserve has entered the pre-FOMC meeting blackout period ahead of the Oct 28 policy meeting, with no official remarks. This week’s asset pricing will hinge more on tech giants’ earnings and economic data: Meta and Tesla will release results, while the semiconductor and storage sectors also enter their earnings disclosure window. Friday’s preliminary PMI readings for the US and Europe, plus the final Michigan consumer confidence index, will test the interest rate hike path. Geopolitically, Trump has suspended full-scale sanctions and tariffs on Russia, setting a roughly 60-day review period. Disruptions to shipping in the Strait of Hormuz and rising Middle East tensions continue to support oil price premiums. Key events this week: Tuesday: NVIDIA GTC Berlin kicks off (runs through Oct 22) Wednesday: US API and EIA crude oil inventories for the week ended Oct 16; NVIDIA CEO Jensen Huang delivers a keynote at GTC Berlin in the evening (Beijing time); US stocks enter a heavy earnings day, with Meta, Tesla, IBM, and Lam Research reporting results after market close Thursday: US initial jobless claims for the week ended Oct 17; Nokia reports earnings before the US market opens, while Western Digital and MaxLinear report after close Friday: Preliminary S&P Global US manufacturing and services PMI for October; final University of Michigan consumer confidence index and 1-year inflation expectations for October; September CPI and preliminary composite/manufacturing/services PMI for Japan, as markets gauge price trends ahead of the Bank of Japan’s Oct 30 policy meeting

HIP-3 market deployer Paragon announced that Papertrade (PAPER) perpetual contracts have launched on the Hyperliquid HIP-3 market, with a maximum leverage of 3x. PAPER is currently trading at $0.012.

Paragon, the deployer of the HIP-3 market, announced that Papertrade (PAPER) perpetual contracts have launched on the Hyperliquid HIP-3 market, with a maximum leverage of 3x.

Beating AI Express, Anthropic has revealed that approximately 90% of Claude Code’s code is written by Claude. From June to September this year, the number of pull requests (PRs) merged by the Claude Code team more than doubled, while the proportion of PRs needing reversion dropped by over 40%. To keep pace with AI’s coding speed, Anthropic has revamped its code review process: before each commit, Claude runs tests first and has independent sub-agents inspect changes; after submission, multiple agents hunt for bugs in parallel, followed by cross-validation from another set of agents. Security reviews and automated tests also block problematic code from merging. Human reviews are tiered by risk: low-risk changes like documentation updates from trusted developers can be merged directly after passing automated checks; regular product modifications require one human approval, while high-risk code changes need two. AI approval cannot replace human oversight, and developers remain responsible for the final code. However, AI reviews can get stuck in loops. Anthropic disclosed that one Claude Code modification went through over 50 rounds of review and fixes overnight. The team found about 59% of the issues stemmed from code that existed during the first review, not new problems introduced by later changes. The team subsequently adjusted rules: if a change goes through more than three rounds of review, the author must reevaluate their approach, and repeated feedback is limited to prevent agents from getting stuck in back-and-forth loops.