10 #Bitcoin ETFs outflows 2,152 $BTC(-$201.21M) and 9 #Ethereum ETFs outflows 45,684 $ETH(-$148.11M) yesterday.

According to official Taiko announcements, Proposal #39 for Ethereum Layer 2 network Taiko has passed the community veto window and been implemented on both Ethereum L1 and Taiko L2. The upgrade primarily optimizes the Taiko Bridge: after the update, the bridge’s ERC20Vault now supports Permit and Permit2, eliminating the need for users to initiate separate Approve transactions when bridging tokens and simplifying the token bridging process. Meanwhile, the Ether gas payment cap for the Taiko Bridge has been raised to proactively adapt to changes brought by EIP-8037. Taiko has also deployed a new BridgedERC20V2 template on both L1 and L2. The project stated that the relevant functions have completed end-to-end testing, covering ETH, ERC20, ERC721, and ERC1155 assets, with bidirectional bridging scenarios also tested.

Beating AI Insight News Brief: OpenAI has added Tailscale support to Codex Cloud, enabling its cloud-based programming agents to access development services on users’ local devices and corporate internal networks. Previously, while Codex could read GitHub code and modify projects in the cloud, it couldn’t directly access resources like APIs and test servers restricted to private networks. Now, users can connect Codex Cloud to their private networks via Tailscale. For instance, if you run a test service on your Mac, Codex can directly call it after modifying code to check results and troubleshoot issues. Enterprises can also grant Codex access to internal APIs, test environments, and private package repositories without exposing these services to the public internet.

Ripple is providing financing for leveraged equity ETFs via its subsidiary Ripple Prime, earning financing fees from related transactions as it enters a business segment long dominated by large Wall Street banks and securities firms. Ripple acquired multi-asset prime broker Hidden Road for $1.25 billion in 2025, gaining capabilities in trade clearing, financing investment positions, and handling transactions in stocks, bonds, foreign exchange, and digital assets. Now, Ripple Prime is offering funds leveraged exposure to stocks and indices through Total Return Swaps, allowing funds to amplify the daily price movements of individual stocks like NVIDIA and SanDisk, while Ripple collects financing fees from these deals. For example, the Tradr 2X Long SNDK Daily ETF pays Ripple an overnight bank financing rate plus 4 percentage points; at current rates, the annualized financing cost is roughly 8%. There are currently 593 leveraged ETFs in the U.S. market, with assets under management exceeding $256 billion. CoinDesk points out that as bank capital and risk regulatory requirements tighten, non-bank institutions including Ripple Prime, Jane Street, and Clear Street are gaining more opportunities to enter this market.

Tom Lee stated at Singapore’s Token2049 conference that BitMine has set a hard cap of 5% of Ethereum’s circulating supply and will stop buying ETH once this threshold is hit. “This is a hard cap. We will not accumulate beyond 5%. We will not hold more than 5% of Ethereum. We only need to purchase around 100,000 more ETH to reach 5%, then we’ll stop,” he said. According to the company’s latest disclosure (as of its most recent purchase on Monday, October 4), BitMine holds approximately 6,016,414 ETH, accounting for about 4.9% of the circulating supply of roughly 122.1 million ETH, leaving a gap of roughly 90,000 to 100,000 ETH to hit the 5% cap. At last week’s buying pace, it will take 6 to 7 weeks to reach the hard cap. Tom Lee also noted that after halting ETH purchases, the company will no longer need to secure financing for buying the asset, which could give shareholders a relative advantage when Ethereum prices rise.

According to Bitget market data, spot silver has fallen more than 2% on the day, currently trading at $58.56 per ounce.

Beating AI News (Insight) — Shenshu Technology has launched its new flagship video model, Vidu Q4 Preview, which is now accessible via web interface and API. The model currently supports image-to-video and reference-based video generation, but does not yet support text-to-video. On Artificial Analysis’s blind test leaderboard for image-to-video (including audio), Vidu Q4 Preview scored 1179 points to rank 3rd, trailing only MiniMax H3 Max and MiniMax H3. Its predecessor, Vidu Q3 Pro, scored 1056 points and ranked 19th, marking a 123-point gap between the two generations. Key improvements of Vidu Q4 Preview include character expressions, limb movements, shot transitions, and visual effects such as explosions. The model can generate both visuals and audio simultaneously, with each video clip up to 16 seconds long and supporting a maximum 4K resolution. Users can also upload up to 15 reference images and 3 audio references, specifying character appearance, scenes, and sounds to maintain consistency of characters across different shots.