Xiaoliu BOT
中EN
Briefs/2026.10.31 Sat
Issue 213

X Platform October 31 AI Brief | Google’s Space AI Chip ‘Suncatcher’ Satellite Passes First Test, Anthropic Adjusts Claude Subscription Plans, Apple Integrates Qwen in China

01 Google’s Space AI Chip “Suncatcher” Test Satellite Achieves First Successful Orbital Test, Completing First End-to-End Large Model Inference on an Orbital TPU

Google officially, along with CEO Sundar Pichai, announced that the first prototype satellite of its “Project Suncatcher,” a collaboration with satellite company Planet, has successfully completed its initial orbital AI inference test approximately one month after launch. The ground control center sent commands to the satellite to wake the onboard data-center-grade TPU. After the model completed the prompt inference calculation, it successfully transmitted back to Earth a haiku about the space environment, marking the project’s successful verification of hardware survival and basic computational power links in the extreme space environment.

According to official announcements and technical blogger summaries, the satellite was launched in early October via a SpaceX rideshare mission. It is equipped with 4 data-center-grade TPUs, aiming to validate the feasibility of powering high-power-consumption AI data centers using the nearly uninterrupted, high-intensity solar energy available in low Earth orbit. In this test, the onboard Gemma 4 model successfully powered on, loaded, and completed its computations, transmitting the results back to Earth as it passed over the Svalbard ground station in Norway. Google stated that the next phase plans to launch two satellites with interconnectivity capabilities in 2027 to further test cross-satellite distributed AI collaborative computing. Future challenges that still need to be addressed include heat dissipation in a vacuum without airflow, protection against high-energy particle radiation in space, and ensuring long-term stability in environments with extreme thermal cycling.

02 Anthropic Adjusts Claude Subscription Plans: Max 20x Renamed to Ultra and Removes 5-Hour Limit for Verified Accounts

Anthropic has officially announced naming and rule adjustments for its Claude high-tier paid subscription plans: the original $100 per month “Max 5x” tier has been renamed to “Max,” and the original $200 per month “Max 20x” tier has been officially renamed to “Ultra.” This adjustment maintains the same monthly fee and total weekly usage limits. However, for accounts under the Ultra plan that have passed the system’s automatic verification, the previous usage frequency limit that reset every 5 hours has been completely removed.

Anthropic official team members Boris Cherny and Alex Albert explained that this adjustment is effective immediately for the vast majority of existing Ultra subscribers. New users will still be subject to temporary window protection until they complete anti-fraud security verification. Previously, heavy developers frequently encountered the issue of triggering a 5-hour circuit breaker during long, continuous programming sessions (such as running Claude Code) even before their weekly quota was exhausted. Removing this window means users no longer need to be forced to pause and wait. Simultaneously, the removal of “20x” from the name is because that multiplier originally referred to the ratio of the 5-hour window quota. The actual total weekly quota remains at 10 times the Pro baseline. This adjustment aligns Ultra’s usage pattern with competitor strategies at a similar price point.

03 Apple Officially Announces Integration of Alibaba’s Tongyi Qianwen into China-Market Apple Intelligence, Set for Public Beta in November and Launch with iOS 27.3

Apple has officially announced a partnership with Alibaba to integrate the capabilities of Alibaba’s Tongyi Qianwen large language model into Apple Intelligence for the China region. Users in mainland China will be able to seamlessly access Qianwen’s text and image understanding, as well as multimodal content generation capabilities in daily use, marking the official, compliant localization of Apple’s AI ecosystem in a major regional market.

Multiple tech bloggers following up with hands-on test information point out that the China-market Apple Intelligence will begin public beta testing on November 4th alongside the latest iOS developer beta version and is planned for a full rollout in December with the official iOS 27.3 release. After integration, Qianwen will be deeply embedded into iPhone’s system-level interaction entry points: users can directly invoke Qianwen within Siri conversations to parse complex intents or understand photos and documents, generate various types of copy and materials via “Compose with Qwen” in the system’s native writing tools, and directly use the camera for image-based queries in visual intelligence scenarios. Community analysis suggests that the core features in the initial launch will primarily focus on highly practical scenarios such as system writing tools, image and text natural language search, real-time multilingual translation, and call/recording transcription.

04 Google DeepMind Releases Real-Time Interactive World Model Genie 4

Google’s AI research arm, Google DeepMind, has officially launched the next-generation foundational world model Genie 4. This model enables users to generate, perceive audio, and interactively control three-dimensional dynamic virtual simulation environments with extremely low latency, pushing multimodal generation technology from static video clips to audible, movable, and controllable real-time digital sandboxes.

Google’s official statement indicates that Genie 4 has broken through the limitations of previous world models that unidirectionally generate pre-rendered videos, allowing intelligent agents or human operators to continuously provide input and receive immediate environmental feedback in real-time interactions. This provides a high-fidelity synthetic training ground for embodied robots and autonomous AI Agents before entering the real physical world. Currently, Genie 4 is being released as a research preview (Research Preview) to trusted testers, and will gradually be integrated into the Google AI Ultra subscription service’s Project Genie platform, with support for calling through the Gemini API and Google AI Studio to the early developer ecosystem.

05 Grok Bot Expands Screen Perception Interaction and Domestic Payment Channels, X Platform Launches Throttling Detection Transparency Feature

xAI’s intelligent agent product Grok Bot and social platform X have welcomed multiple interaction and platform function updates. Grok Bot recently not only enabled multimodal Q&A capabilities by long-pressing the mobile shortcut key to directly read the current screen content, but was also discovered by domestic developers to have officially integrated Alipay and UnionPay payment settlement channels, significantly lowering the subscription threshold for users in non-dollar regions; meanwhile, X officially announced the launch of a tweet throttling detection label, enhancing the transparency of algorithmic risk control.

According to Elon Musk’s cited discussions and community actual experiences, Grok Bot possesses the ability to directly view videos playing on the phone screen, social dynamics, or lengthy web articles and extract key information. Musk emphasized that such intelligent agents should be regarded as closely cooperating “work partners.” At the platform transparency level, X’s official open-source and technical team added a restricted status display function in the “Under the Hood” panel. When less than 5% of the content in a user’s published tweet is subjected to visibility-limiting labels by the system, creators can directly and clearly view in the backend which specific tweet was throttled, providing a more intuitive self-check basis for social media algorithm rules.

06 Europe Discusses Data and AI Taxation: French National Assembly Preliminary Review Passes Internet Traffic Tax Budget Amendment Sparking Widespread Controversy

The French National Assembly, with a narrow margin of 11 votes in favor and 10 votes against, preliminarily reviewed and passed an internet traffic tax amendment to the 2027 budget bill. This proposal intends to impose a tax of 0.055 euros per 1GB of data transmitted to content and service providers within French territory, with an estimated annual collection scale of 10 billion euros; simultaneously, discussions have emerged in German political circles regarding the imposition of taxes on all economic value created by AI, attracting high attention from the tech venture capital and open-source developer communities.

Technical community developers and practitioners generally express concerns about the chain reaction of this bill. Analysis indicates that streaming media, cloud service providers, and AI Agents that frequently transmit large multimodal files will all face direct operational cost impacts; especially for the open-source model ecosystem, similar to Hugging Face hosting hundreds of GB of open-source weights, if taxed per GB, a single download could incur tax burdens of tens of euros, inevitably passing the costs onto end developers and consumers. Although this amendment is still far from formal legislative implementation and requires lengthy legislative and judicial review procedures, multiple observers in the European and American tech circles point out that this reflects some traditional economies, when facing rapid AI development and public fiscal austerity, are accelerating the exploration of new tax regulatory tools targeting digital data transmission and computing power achievements.

07 DeepSeek Harness Community Plugin Ecosystem Expands: Multiple Developers Launch Design and Workflow Suites

The local development and extension ecosystem around DeepSeek Harness (DSH) is rapidly gaining momentum within the technical community. Several well-known developers have successively open-sourced and released practical plugins for design orchestration, kanban management, and multi-agent collaboration, driving the Agent toolchain from single-command execution towards systematic coordination.

Developer Baoyu (@dotey) launched the baoyu-design extension, enabling Claude Design to be seamlessly embedded into DeepSeek Harness for operation, and introduced features for page and slide element tagging and contextual comment linkage. Designer blogger Guizang (@op7418) consolidated their previously scattered PPT, social card, and web typography components optimized for Chinese typesetting into a unified “dsh-guizang-design” comprehensive plugin. This plugin not only shares the underlying design system and Design Tokens but also includes built-in visual generation kanban and adaptive style imitation modules. Independent developer Tony Dinh also published a discussion exploring the practical experience of building multi-agent automated generation interfaces based on the Harness architecture. Feedback from various parties indicates that integrating design specifications, multi-model scheduling, and review processes into a closed loop through custom Harness configurations is significantly enhancing automation efficiency in high-frequency production scenarios.

← All briefs →