Binance will delist the USDC spot trading pairs of AUCTION and VANA.

Belarus has granted registration approval to its first two crypto banks, which will provide fiat currency deposits, tokenized transfers, and digital asset custody services.

Beating AI Express (from Dongcha) reports that open-source Computer Use tool Cua has added an optional Perception visual extension to its Cua Driver. The Driver originally prioritized locating buttons via web structures and system accessibility trees, but these structures may be empty when dealing with Canvas, games, remote desktops, or custom-drawn applications. Perception directly reads screenshots, splitting the screen into regions with text, type, and position for the Agent to operate on. The first version uses OmniParser to identify icons and controls, and PP-OCR to read text. The entire parsing process runs on the local CPU, no GPU required, and screenshots are not transmitted over the network. Official tests show processing a single screenshot takes roughly 2–2.5 seconds on macOS, 3.5–4 seconds on Linux, and 8–9 seconds on Windows. It acts as a fallback when page structures and accessibility trees fail, and does not run visual parsing by default for every step. Cua also restricts the reuse of old screenshots for visual operations: each visual recognition is bound to the current screenshot’s `capture_id`, and subsequent clicks must originate from the same screenshot. If the screenshot is over 60 seconds old, or an operation has already been performed, the Driver will refuse to continue using it, preventing accidental clicks with outdated coordinates after the interface changes. Cua was already capable of visual positioning via `cua-som` and OmniParser in the past; now this functionality is directly integrated into Cua Driver, allowing upper-layer Agents to obtain recognition results via a unified interface. Even if the model itself cannot process images, it can decide on the next step based on these text and region data. Perception is not a default component: the OmniParser detector uses the AGPL-3.0 license, while Cua Driver remains under the MIT license.

Former U.S. Commodity Futures Trading Commission (CFTC) Chair Rostin Behnam told Crypto In America in an interview that with the explosive growth of prediction markets, Congress urgently needs to clarify the boundary between risk management and gambling, and related lawsuits may ultimately be brought before the U.S. Supreme Court. Behnam also noted that after the Clarity Act failed to advance in the Senate, the CFTC’s reliance on existing authority to regulate spot markets is akin to “patching with tape and paper clips,” adding that regulators urgently need to craft more durable rules.

Beating AI Insight News Brief: Anthropic has added an automatic tuning workflow to Claude Code’s official claude-api Skill. The build-eval module first generates test sets and scoring metrics from real conversations, bugs, or manual cases; hillclimb then enables Claude Code to iteratively modify system prompts, tool descriptions, models, and inference intensity. After each adjustment, tests are rerun: improvements are retained, while regressions are rolled back. To prevent Claude from overfitting exclusively to test cases, tuning is not performed on all samples at once. A subset of cases is used to identify issues and adjust configurations, while a separate batch is reserved to validate if modifications work on unseen problems. If the primary score improves but results on the reserved test set do not, the modification may be deemed overfitting and withdrawn. Anthropic demonstrated this workflow using 44 customer support tickets. Claude Code sequentially cleaned prompts, added business rules, tested different models, and reduced inference intensity, ultimately settling on a low-effort configuration for Sonnet 5. On 14 tickets not included in tuning, accuracy rose from 78.6% with the initial configuration to 90.5%, while costs fell to roughly one-fifth of their original level.

South Korea's Kakao Pay Securities has partnered with Ondo Finance and Dinari respectively to explore tokenizing South Korean-listed stocks and providing overseas investors with on-chain investment channels. Kakao Pay Securities is one of South Korea's leading online securities firms, backed by the Kakao financial ecosystem. As of the first half of 2026, its client assets under custody exceeded 21 trillion won (approximately $150 billion), with around 1.61 million monthly stock trading users. The collaboration aims to research integrating South Korean stock assets onto the blockchain, boosting the accessibility of traditional securities assets via tokenization and creating a new channel for overseas investors to access South Korea's stock market.

Beating AI Express News: X is testing a new unified subscription plan. An early Android test page leaked by the community features an "X Subscription" upgrade card, which states that a single plan will include access to Cursor, Grok, and X’s core benefits, with a Grok Bot icon displayed at the top. Maxime Prades, head of SpaceXAI’s product team, previously confirmed the company is developing a "more unified subscription" that will roll out in "weeks, not months." Currently, several subscription tiers have partial integration but remain charged separately: Cursor Pro, Pro+, and Ultra all include Grok Bot access; SuperGrok, SuperGrok Plus, Heavy, and X Premium+ also grant Grok Bot benefits to Cursor accounts. However, Cursor’s official statement clarifies that this linking does not convert Grok or X subscriptions into Cursor plans, nor does it cancel existing Cursor subscriptions.