Meta Integrates AI Model Muse Into A Keychain Device: Users Can Chat Via Fingerprint Press, With Shipping Scheduled For December.

According to Arkham’s monitoring, a certain address received 45,000 ETH from Galaxy Digital’s OTC desk. At the time of the transfer, the sum was valued at approximately $119.67 million, with an average entry price of around $2,659 per ETH. The address then combined some of its existing holdings with the newly received funds, routed them through intermediate wallets, and transferred a total of roughly 54,990 ETH to the same destination wallet. As of press time, this aggregated wallet holds about 57,257.78 ETH, worth approximately $153.4 million, with no active outgoing transfer records. Aggregation address: 0x0ff2c6e74e97c4c229bf156ab1ed8dd97bc84c9f

Beating AI Breaking News: Tencent’s QClaw to Cease Operations on December 24 Tencent’s local AI agent QClaw announced it will officially shut down on December 24. Effective immediately, new user registration, subscription purchases, and renewals are halted. Existing users can still access the service, and unused paid subscriptions and point packages are eligible for refunds. Post-shutdown, only data backup and migration functions will remain; starting March 24, 2027, all user data not required by law to be retained will be deleted. Developed by Tencent’s PC Manager team based on OpenClaw, QClaw launched internal testing on March 10 this year. Its core function is enabling users to remotely command computers via WeChat to process files and complete tasks. In April, Tencent expanded QClaw to overseas markets. At the June Tencent Cloud AI Conference, QClaw and WorkBuddy were still promoted as two distinct personal productivity AI agents. The restructuring occurred in July: Tencent integrated QClaw’s product center business and part of its team into Cloud Product Department 6, which oversees WorkBuddy. An internal notice at the time claimed “QClaw will continue operations”. Just two months later, the shutdown was announced. Tencent has added a “Migrate to WorkBuddy” portal on QClaw’s official website. Users can transfer configurations, memories, AI agent personalities, and conversation histories; those who complete a valid migration will receive 1,000 points.

Polygon Foundation CEO Sandeep Nailwal announced in a post this morning that the burning of 100 million POL tokens—roughly 1% of POL’s total supply—has been completed. Earlier, Nailwal noted that Polygon’s cumulative base fees have accumulated to 12.1 million POL tokens. He added that POL has been in a deflationary state since January 2026, with network activity continuing to grow and the network scaled to 5,000 TPS.

UK’s Financial Times, citing recent court documents, reports that after Apple integrated ChatGPT into Apple Intelligence, OpenAI found the number of new weekly active users fell short of expectations, prompting it to revise down its growth forecasts. By summer 2025, OpenAI internally viewed this integration as “clearly falling short of expectations.” The documents show ChatGPT is set to default off in Apple Intelligence, requiring users to manually enable it, a factor that may have limited its actual usage. While users can access ChatGPT via Siri and the system’s writing tools, Apple controls the entry point for AI features, how they are accessed, and subsequent model selection. OpenAI had previously expected to leverage iPhone’s massive user base to expand ChatGPT’s reach and drive more users to convert to paid subscribers, but actual subscription conversions also fell short of projections. The documents do not disclose specific figures for new users or subscriber counts. This year, Apple chose Google Gemini as the model base for its new version of Siri AI, further highlighting Apple’s control over core AI entry points. The related disclosures come from court documents in a lawsuit between Apple and OpenAI, reflecting OpenAI’s stance in the litigation, not the public operating data released by either company.

Binance announced that it will support the Injective (INJ) network upgrade and hard fork, and will suspend INJ token deposits and withdrawals starting at 22:00 Beijing time on September 24. The upgrade is scheduled to take place at block height 184,394,000, and is expected to complete around 23:00 Beijing time on September 24. INJ spot trading will remain unaffected during the upgrade. Binance stated that it will handle the relevant technical work, and will reopen deposits and withdrawals once the upgraded network stabilizes, with no separate announcement to be made for this process.

Professional data firm Silicon Data published a report noting that the market has recently frequently cited its LLM Token Index as a basis for bearish bets on AI trading, but the index’s decline does not signal a weakening of AI demand or model-layer profits. Silicon Data explained that its LLM Token Index measures the actual price paid per million tokens by API users in its sample — referred to as the "usage-weighted price" — rather than token volume or total spending. The index fell 52% from $2.07 on May 28 to $1 on September 21, but had risen 67% between January and its May peak, meaning price movements alone cannot serve as a direct indicator of AI industry sentiment. The firm pointed out that the recent index decline mainly reflects more tasks being allocated to cheaper, faster models, while AI labs continue rolling out additional mid-tier models. Since tokens do not fully represent AI capabilities, the index also cannot distinguish between model substitution and more efficient agent routing on its own. By contrast, the GPU computing power rental market still shows strong demand. Silicon Data’s data shows that rental prices for H200 GPUs at non-hyperscale cloud providers rose from an average of $2.87 in June to $3.29 currently, hitting $3.32 on September 19; B200 GPUs now trade at $5.76, up 7% year-over-year and 31% year-to-date; B300 GPUs have risen 44% since their launch at the end of April. Silicon Data believes that with the large-scale adoption of agents, the vast majority of tokens in the future will likely come from cheap, fast models, so token prices can continue to decline while total token volume still grows significantly. The firm added that truly high-value tasks will likely remain concentrated on frontier models, and the widespread adoption of AI means demand for computing power could be "more, not less".