Anthropic files for an IPO, planning to list on Nasdaq under the ticker “ANT”
Starting at 5,499 yuan, the iPad mini 8 has been released, but it still lacks a high refresh rate.
Apple responds to iPhone Duo internal screen cracking: A small number of devices are affected; customers can contact AppleCare for a replacement.
OpenAI releases GPT-6.1 Terra, with API prices cut in half and faster computer performance.
Apple officially releases iOS 27.2: Apple Music gets a new black-and-white design, and iPhone Duo now supports the Apple Pencil.
Huawei’s Enjoy series has cumulatively shipped over 160 million units, and the Enjoy 100 will be launched on October 28.
Xpeng Youdao has been approved for L4-level demonstration operations, with initial access still limited to users holding an invitation code.
DeepSeek Flash 1017 tops the OpenRouter weekly usage chart, processing 33.6 trillion tokens.
SK Hynix responds to rumors of a special dividend: the plan remains undecided, and it intends to announce its shareholder return strategy on the 28th.
Kimi apologizes for the rate limiting on K3.1 Pro and resets the quota.
Audi China has clearly defined its brand divisions: FAW focuses on the four-ring logo, while SAIC concentrates on the AUDI lettering.
Volkswagen China has obtained an L3 road-testing license and has already launched real-world road tests in Hefei.
Anthropic confirmed that some internal systems were accessed by researchers, but customer conversations and model weights were unaffected.
Cadence launches the Maestro chip design model, which can run within the customer’s firewall.
Microsoft is offering a voluntary severance package to approximately 60,000 employees, with decisions to be made by early December at the latest.
Meta launches approximately $35 billion in equity financing and also plans to introduce a $60 billion share sale program.
SpaceX has secured approximately $40 billion in GPU financing commitments and will lease NVIDIA’s Vera Rubin equipment.
Bill Gates: Before AI brings abundance, we must first endure about 20 years of upheaval.
Lu Weibing claims the REDMI K100 is “unrivaled in its class,” and the new Snow Rock White colorway has been unveiled.
WEY Mocha 8 HEV launched, with a limited-time trade-in offer starting from RMB 259,800 for the plug-in hybrid version.
Wuwen Xinqiong and the Three-Body Universe have launched the Zhi Zi Agent, which can maintain cross-session memory.
Tencent Hunyuan has open-sourced Hy-Nomas, which can generate video worlds segment by segment using both text and camera footage.
Ji Meng has launched a short‑video studio, enabling scripts, storyboards, and localized reshoots to be completed on the same platform.
Tencent has open-sourced the Chinese–English simultaneous translation model Hy-SimulMT, which supports real-time listening and translation as well as voice tone cloning.
Lao Gan Ma denies “reducing staff from 600 to 48,” confirming that AI-based quality inspection has been implemented.
ROG × “Shadow Blade Zero” first batch of co-branded hardware is now on sale, with martial-arts-inspired designs spanning both gaming consoles and peripherals.
Dianping’s internal beta of “Life Expert”: You can earn growth points by finding stores, checking in, and interacting.
On its fourth day of release, “Sanxingdui: A Future Past” grossed approximately RMB 1.19 million, while Bona Film Group’s stock hit the daily limit down.
The 2026 “Harmony Starlight Gala” is scheduled for December 5.
Lu Yang’s produced film “Probe” is set for release on November 21, with a new trailer unveiled.
Anthropic files for an IPO, planning to list on Nasdaq under the ticker “ANT”
On October 26, Anthropic publicly filed an S-1 registration statement with the U.S. Securities and Exchange Commission, planning to list on the Nasdaq under the ticker symbol “ANT.” The company stated that the number of shares to be issued and the price range have not yet been determined, and the registration statement has not yet become effective.
The prospectus disclosed that revenue for the first half of this year was $16.286 billion, compared to $1.163 billion in the same period last year; GAAP net loss amounted to $80.748 billion. Notably, changes in the fair value of convertible bonds, derivative liabilities, and future equity simple agreements resulted in a loss of $72.388 billion, primarily attributable to an increase in the company’s valuation and not equivalent to cash burn during the same period.
Operating losses amounted to $693 million. Excluding items such as share-based compensation and non-cash charitable donations, adjusted operating profit was $1.618 billion, reported under non-GAAP measures.
The document also notes that future revenue growth may slow, and the expansion of computing power and infrastructure will continue to put cost pressures on the company; highly autonomous AI systems, if misused or lacking adequate safeguards, could also pose catastrophic risks. The special Class F shares held by the company’s founding entities will carry 50.1% of the voting rights on matters requiring shareholder approval.
Related reading: Breaking news! Anthropic has filed for an IPO—could this be the largest IPO in history?
Starting at 5,499 yuan, the iPad mini 8 has been released, but it still lacks a high refresh rate.
Last night, Apple quietly launched the new iPad mini on its official website, featuring an 8.5-inch OLED Super Retina XDR display and the A20 Pro chip. The device is just 5.4 mm thin, and the Wi‑Fi model weighs 255 grams. Despite the screen upgrade, it still does not support the 120Hz ProMotion high refresh rate.
The main changes in the new device include:
The display bezels have been narrowed by 24%, with HDR peak brightness up to 1,200 nits and a minimum brightness of 1 nit; the 256GB and 512GB models offer optional nano-textured glass.
The A20 Pro uses a 2-nanometer process and is equipped with a 7-core GPU across all variants; the 128GB and 256GB models feature a 5-core CPU and a single 16-core neural network engine, while the 512GB model is upgraded to a 6-core CPU and a dual 16-core neural network engine.
It adds an Acoustic Display audio system and supports IP67-rated water and dust resistance; the front-facing 12-megapixel Center Stage camera has been moved to the longer edge. The N1 chip supports Wi‑Fi 7, Bluetooth 6, and Thread, while the cellular version is equipped with a C2 baseband.
Supports fast charging; when paired with a separately purchased 35W or higher‑power compatible adapter, it can charge up to 50% in about 30 minutes.
The new model is available in Midnight and Starlight colors, with the 128GB Wi‑Fi version starting at RMB 5,499 and the cellular version starting at RMB 6,799. Pre-orders begin on October 28, with the official release on November 6. Also launched simultaneously is the Tech Fabric Smart Bifold Case, priced at RMB 599, which supports both the Apple Pencil Pro and the USB‑C‑based Apple Pencil.
Related reading: New iPad mini unveiled: starting at 5,499 yuan! It has everything you could want—except high refresh rate.
Apple responds to iPhone Duo internal screen cracking: A small number of devices are affected; customers can contact AppleCare for a replacement.
Apple has responded for the first time to complaints about cracked inner screens on the iPhone Duo. In a statement sent to the media, Apple said it has become aware of issues with the inner screens of a small number of iPhone Duos and advised users experiencing display abnormalities to contact AppleCare to arrange a device replacement.
We have learned of an issue affecting the internal screens of a small number of iPhone Duo devices. If a user’s iPhone Duo display is experiencing problems, they should contact AppleCare to arrange for a replacement device.
Apple also stated that the vast majority of devices are unaffected, and that faulty models typically exhibit issues early in use.
According to Apple employees, the company has not yet issued formal guidelines to retail stores or customer service staff, but some teams have already launched investigations, and certain engineers are working to retrieve affected devices for analysis.
Previously, some users posted photos on forums like Reddit showing the inner screen cracking along the hinge’s centerline. iFixit’s teardown suggested that components might be exerting excessive pressure on the display.
OpenAI releases GPT-6.1 Terra, with API prices cut in half and faster computer performance.
OpenAI has released GPT-6.1 Terra, which focuses on visual understanding, computer operations, and natural writing. It is now available to Plus, Pro, Enterprise, and Edu users of ChatGPT Work and Codex, and developers can access it via API calls.
OpenAI stated that the new model’s standard API pricing is 50% lower than GPT-5.6 Terra and GPT-6.1 Sol: $1 per million input tokens, $0.05 for cached inputs, and $5 for outputs.
In the “Last Exam” test, which includes specialized tasks such as converting architectural blueprints into 3D models, Terra outperformed the previous generation by 15 percentage points at medium reasoning intensity, achieving results comparable to GPT-6 Astra; according to official statements, its speed in completing computer-based tasks exceeds that of GPT-6 Astra and GPT-6.1 Sol by a factor of three.
In terms of programming, Terra achieved a 50.3% pass rate on Terminal-Bench 4.0 at maximum reasoning intensity, compared to the previous generation’s public score of 21.5%. OpenAI continues to prioritize Sol and Astra as the most complex programming tasks.
Related reading: Just now, GPT-6.1 Terra was released at a discounted price—here comes the打工 model for the Agent era.
Apple officially releases iOS 27.2: Apple Music gets a new black-and-white design, and iPhone Duo now supports the Apple Pencil.
Early this morning, Apple officially rolled out iOS 27.2, which includes design tweaks to several native apps and expands Siri’s language support.
In terms of visual and interface design, the Apple Music icon has been updated from its traditional red color to a black-and-white design, and the accent color for selected tabs has been changed from red to blue. Additionally, a music widget with a black background has been introduced.
This update also adds support for the Apple Pencil (USB-C) on the iPhone Duo.
On the functional level, Siri’s AI language capabilities have been further expanded, with support now added for five additional languages: French, Japanese, Korean, Portuguese, and Spanish.
Other native apps and features have also received corresponding improvements.
The Messages app has added a “Suggested Actions” feature that automatically provides shortcuts—such as creating a calendar event or setting a reminder—based on the conversation content.
The Podcasts app separates “New Episodes” and “Now Playing” on the home screen.
Apple TV has added support for multiple user profiles;
The dual-camera recording feature of Group FaceTime has been extended to group calls, allowing users to share footage from both the front and rear cameras during a call (unavailable in mainland China).
In the smart home domain, compatible smart locks now include the “auto-lock after closing” and “reminder for doors left open too long” options, and HomePod, following its update, also supports “lock control.”
Huawei’s Enjoy series has cumulatively shipped over 160 million units, and the Enjoy 100 will be launched on October 28.
On October 26, He Gang, a Huawei board director and CEO of the Consumer BG, stated in a post that Huawei’s Enjoy series has reached its 11th anniversary, with cumulative shipments exceeding 160 million units. He cited durability, reliability, and long battery life as key reasons why users choose this series, and expressed gratitude to users for their continued support.
The new-generation Enjoy 100 series will be unveiled on October 28 at 14:30. The figure of 160 million units announced this time represents the cumulative shipments across the entire Enjoy series, spanning multiple generations of products—not the order volume or sales figures for the Enjoy 100 alone.
Xpeng Youdao has been approved for L4-level demonstration operations, with initial access still limited to users holding an invitation code.
Xpeng Robotaxi has obtained Guangzhou’s qualification for demonstration operations of intelligent connected vehicles and is now authorized to conduct L4‑level demonstration services throughout the entire city. On October 26, NanDu·WanCaiShe learned from Xpeng Group that, in the initial phase, “Xpeng YouYou” will continue to roll out access gradually via invitation codes while steadily expanding its service coverage.
This marks a step forward in gaining approval for real-world mobility services, following road tests and internal employee trials. In May this year, XPeng’s first mass-produced Robotaxi rolled off the assembly line; in July, it completed end-to-end operations, from online booking to automated pickup and delivery; and in August, it obtained authorization to conduct road tests in Guangzhou without a safety driver in the driver’s seat. In October, the service was officially named “XPENG YOYO,” and its ride-hailing mini-program went live simultaneously.
Xiaopeng Youyu is built on the Xiaopeng GX and is equipped with four self-developed Turing AI chips and a second-generation VLA large model. Earlier, He Xiaopeng stated that the business model aims to provide partners with a full-vehicle platform, autonomous driving software, and AI capabilities, while partners handle local fleet operations; revenue will come from hardware sales, technical services, and commission sharing.
DeepSeek Flash 1017 tops the OpenRouter weekly usage chart, processing 33.6 trillion tokens.
OpenRouter’s most recent weekly usage ranking as of October 25 shows that DeepSeek V4.1 Flash 1017 processed 33.6 trillion tokens, an 801% month-over-month increase, placing it in first place.
The second through fifth places on the list are, in order, DeepSeek V4.1 Flash, MiniMax M3.1 Flash Free Edition, GLM 5.3 Flash, and Tencent Hy4 preview, with processing volumes of 18 trillion, 13.7 trillion, 13.2 trillion, and 8.97 trillion tokens, respectively.
This ranking lists the input and output tokens processed by the OpenRouter API, excluding requests that users or applications have marked as private. Different variants of the same model are ranked separately. It reflects usage within the platform, does not include direct services from vendors worldwide, and is not a ranking of accuracy or inference performance.
OpenRouter also tallies vendor market share based on the number of text requests: in the week beginning October 19, DeepSeek accounted for 26.5%, again ranking first. The share of request counts and the total token volume are two distinct metrics.
SK Hynix responds to rumors of a special dividend: the plan remains undecided, and it intends to announce its shareholder return strategy on the 28th.
According to South Korea’s Edaily, SK Hynix responded on October 26 to rumors of a special dividend, stating that the company is exploring various shareholder-return strategies but has not yet made a final decision. It plans to disclose any new arrangements when it releases its third-quarter earnings report on October 28, following a board resolution.
Previously, some South Korean media reported that SK Hynix is considering using the proceeds from the sale of its stake in Japanese memory manufacturer Kioxia to distribute a special dividend of approximately 16 trillion won.
SK Hynix had previously announced a share buyback and full cancellation plan totaling 40 trillion won. According to Edaily, the buyback was completed ahead of schedule on October 21, and all shares repurchased will be canceled.
Kimi apologizes for the rate limiting on K3.1 Pro and resets the quota.
Kimi yesterday issued a public apology for the throttling and unnecessary quota consumption in the K3.1 Pro subscription plan, and announced a one-time reset of usage for all subscribers across all plans.
Kimi officially stated on X that these issues “were on our end” and should have been fixed earlier. The compensation plan is to restore 25% of each subscriber’s monthly quota across all plans, equivalent to one week’s usage.
For users currently below 25% of their allocated usage, the balance is reset directly to zero. Kimi also stated that it has adjusted its plan parameters, reducing unnecessary sub-agent calls and output lengths to enable the same quota to accomplish more real-world tasks.
This incident occurred two days after the release of K3.1 Pro. One of the key upgrades of this model focuses on financial research; internal testing by Kimi shows that its financial modeling score improved from 50.4 in K3 to 57.6, and its financial research score rose from 58.8 to 70.4.
Audi China has clearly defined its brand divisions: FAW focuses on the four-ring logo, while SAIC concentrates on the AUDI lettering.
Audi’s two brand systems in China now have a clearer division of responsibilities. On October 26, according to First Financial reporter Wu Ziyue, Audi signed strategic agreements separately with China FAW and SAIC Motor, clearly defining the differentiated positioning of the four-ring brand and the AUDI letter‑logo brand, as well as a “dual-brand, dual-partner” channel structure.
China FAW will strengthen Audi’s four-ring brand positioning across both the intelligent luxury new-energy vehicle segment and the conventional fuel-powered vehicle segment; SAIC Audi, on the other hand, will focus on the AUDI lettering brand, concentrating on the new-energy market and developing the next-generation smart connected vehicles tailored for Chinese consumers.
During the transition period, SAIC Audi will continue to sell its four-ring products as usual until the end of 2027. Both the four-ring models under SAIC Audi and the after-sales service for AUDI-branded vehicles will remain in stable operation.
Volkswagen China has obtained an L3 road-testing license and has already launched real-world road tests in Hefei.
Volkswagen Group China announced that it has obtained an L3-level autonomous driving road test license in Hefei and has commenced real-world road testing and data collection.
Volkswagen China announced on October 26 that this licensing represents a step toward achieving L3-level automated driving technology readiness by 2027. The scope of the authorization is limited to road testing, and the group has set 2027 as its target for technological readiness.
On October 28, the inaugural Volkswagen Group (China) Technology Day will be held in Hefei, showcasing the latest technological advancements.
Anthropic confirmed that some internal systems were accessed by researchers, but customer conversations and model weights were unaffected.
According to a report by The Washington Post on October 26, the cybersecurity startup Cathedral exploited vulnerabilities in third-party enterprise software under Anthropic’s external security testing program to gain access to its internal systems. A draft of Cathedral’s report stated that the AI agent, after running for less than 12 hours, identified pathways to access meeting transcripts, compensation and hiring information, as well as Slack messages related to certain discussion models.
Anthropic confirmed that researchers accessed limited portions of its sales, human resources, and IT systems, including some customer and partner business contact information as well as basic account records; customer messages sent to the model, model responses, AI system code, and model weights were not accessed.
The company stated that its monitoring system had detected some activity prior to Cathedral’s report and had revoked access that same day; at present, there is no evidence indicating that others exploited these vulnerabilities. The affected enterprise software was disabled and will only be reactivated once the vendor confirms the fixes have been applied.
The two parties still disagree on the testing process. Anthropic argues that Cathedral has exaggerated the national security implications and failed to fully disclose its activities within the system, while Cathedral maintains that it followed standard security research practices by limiting data access to the minimum necessary level.
Cadence launches the Maestro chip design model, which can run within the customer’s firewall.
Chip design software company Cadence has unveiled the Maestro AI model suite to power its chip and system design agents. The models can be deployed on Cadence Cloud, public clouds, or run within customers’ own data centers and behind their firewalls.
Maestro partly leverages NVIDIA’s open-weight Nemotron model and integrates OpenShell security technology; customers can fine-tune it using their own chip design data. Cadence connects the model and reinforcement learning workflows with EDA design tools, continuously optimizing design decisions based on engineering outcomes.
The company stated that, in initial design‑process testing, it achieved up to a 30% improvement in quality and up to five times greater token‑usage efficiency compared with general‑purpose base models; these are the results of its early tests. The first batch of models is scheduled to be made available to select early‑adopter customers in the fourth quarter. Paul Cunningham, Cadence’s senior vice president, told Reuters that Maestro is pursuing both open‑source and closed‑weight model approaches.
Microsoft is offering a voluntary severance package to approximately 60,000 employees, with decisions to be made by early December at the latest.
According to Business Insider, after reviewing the employee plan and receiving confirmation from Microsoft, the company is offering voluntary severance packages to approximately 60,000 employees, covering teams in the United States, Canada, the United Kingdom, and parts of India. The number of eligible employees does not necessarily reflect the final headcount of those who will leave; employees can decide whether to accept by early December and are expected to depart by year-end.
Compensation is calculated separately based on job grade, region, and length of service. According to the U.S. proposal reviewed by this media outlet, employees at Grade 64 and below receive one week’s base salary for every six months of service, while those at Grades 65 to 67 receive two weeks’ base salary per half-year; the maximum standard compensation for Grades 67 and below is capped at 39 weeks of base salary.
The package also includes stock vesting and health insurance, with standard stock vesting periods of six or twelve months—depending on length of service—for employees at Level 67 and below. A Microsoft spokesperson stated that there are currently no plans to launch another voluntary separation program of this kind.
Meta launches approximately $35 billion in equity financing and also plans to introduce a $60 billion share sale program.
On October 26, Meta announced the launch of two public offerings totaling approximately $35 billion, comprising about $22.5 billion in Class A common stock and roughly $12.5 billion in depositary shares representing mandatory convertible preferred shares. The company stated that these arrangements will support investments in AI infrastructure and “personal superintelligence.”
Another share-sale plan valued at approximately $60 billion, denominated at market price, is slated to commence once the newly issued Class C common shares begin trading, with an anticipated launch no earlier than the first quarter of 2027. The combined base size of both components totals roughly $95 billion; as the future plan has not yet been formalized into a binding agreement, the actual timing of commencement, scale, and terms remain undetermined.
NVIDIA, AMD, and Broadcom have each expressed interest in purchasing up to $5 billion worth of common stock in this offering, though these commitments are non-binding and the actual purchase amounts may be adjusted.
This financing will also be received in stages. The Class A common stock offering is structured as a forward sale agreement; Meta will not receive the proceeds from this portion of the shares upfront but will instead obtain the funds upon future settlement. The company stated that it will support its computing infrastructure development through a combination of operating cash flow, investment-grade debt, strategic partnerships, and equity capital.
SpaceX has secured approximately $40 billion in GPU financing commitments and will lease NVIDIA’s Vera Rubin equipment.
On October 26, SpaceX announced that it has entered into a binding commitment to secure approximately $40 billion in master financing for GPUs and related equipment managed by an entity under Apollo Global Management, to expand its ground-based computing infrastructure. The equipment will include NVIDIA’s next-generation Vera Rubin chips.
This financing round was led by Apollo and Goldman Sachs, with Goldman Sachs serving as the exclusive structuring agent; BlackRock, funds and accounts managed by Goldman Sachs Asset Management, and capital accounts advised by KKR also committed to make core investments.
The transaction remains subject to the agreed closing conditions and is expected to close by the end of 2026, with the specific leasing arrangements anticipated to commence following the delivery of the GPUs and related equipment in the first half of 2027.
SpaceX stated that the additional computing power will support Grok for both enterprise and consumer users, as well as existing and future cloud service customers; leasing financing is also part of its strategy to diversify funding sources and accelerate its computing capacity deployment plans.
Bill Gates: Before AI brings abundance, we must first endure about 20 years of upheaval.
In a Decoder podcast interview aired on October 22, Bill Gates stated that AI could help alleviate shortages in areas such as housing, food, and healthcare, but he is more concerned about the roughly 20-year transition period leading to abundance—during which there will be winners, losers, and significant disruptions.
In Gates’s view, white-collar jobs such as telemarketing, customer service, and accounting will be affected first, and humanoid robots may subsequently transform manual labor as well. Work provides income, a sense of purpose, collegial relationships, and opportunities to solve problems; simply making goods cheaper cannot explain how people obtain these benefits.
He cited his own philanthropic work as an example: even though mobilizing resources to eradicate malaria was once his unique contribution, AI might do it better—sending him off to play pickleball instead. For young people who aspire to make a distinctive impact, this shift in capabilities can be stressful.
Gates suggested that humanity could choose to reserve roles such as childcare, cooking, the arts, and sports for humans, even if AI can perform them better; early retirement and a shorter workweek are also viable options. Such arrangements would require society to make policy decisions. He believes that some individuals might deliberately step back from the technological forefront to preserve space for mutual support and for making unique contributions.
He also emphasized that 20 years is merely a relatively conservative estimate he has chosen; the actual timeline could be longer or shorter. Unless AI development is significantly slowed down or halted, robots will gradually take on tasks such as building homes, growing food, and performing surgeries. Humanity will need to decide how to organize life in an era of abundance, while his current focus remains on navigating this period of transition.
Lu Weibing claims the REDMI K100 is “unrivaled in its class,” and the new Snow Rock White colorway has been unveiled.
The REDMI K100 will be unveiled on October 27 at 19:00. On October 26, Lu Weibing, President of Xiaomi Group, posted a teaser, calling it “another unbeatable product in its class.”
On the same day, REDMI officially unveiled a new color option, “Snow Rock White,” featuring a delicate texture on the back of the device reminiscent of Xuan paper, and pre-orders have already opened on major e-commerce platforms.
WEY Mocha 8 HEV launched, with a limited-time trade-in offer starting from RMB 259,800 for the plug-in hybrid version.
WEY Mocha 8 HEV was launched on October 26, offering two trims—the Comfort Edition and the Luxury Edition—with official prices of RMB 265,800 and RMB 285,800, respectively. According to information obtained by Autohome from WEY’s official sources, the limited-time trade-in incentive price is RMB 259,800 for the Comfort Edition and RMB 279,800 for the Luxury Edition.
The new vehicle features a plug-in hybrid system comprising a 2.0T engine and dual front-and-rear electric motors, with Hi4 all-wheel drive standard across the entire lineup. It is equipped with a 3.64 kWh battery and does not require external charging. The system delivers a combined maximum power of 322 kW and a peak torque of 782 N·m, achieving 0–100 km/h acceleration in 5.9 seconds, with a WLTC combined fuel consumption of 6.92 L/100 km.
The vehicle measures 5,280 mm in length with a 3,145 mm wheelbase, featuring a three-row, seven-seat layout that maintains a flat floor and a common rail design for the second and third rows. The cabin is equipped with Coffee OS 3.5 and a Qualcomm 8295P chip, supporting HiCar, CarLink, and CarPlay. The advanced driver-assistance system is Coffee Pilot 3, outfitted with a 192-line LiDAR, 27 sensors, and an Orin‑Y chip.
Wuwen Xinqiong and the Three-Body Universe have launched the Zhi Zi Agent, which can maintain cross-session memory.
On October 25, at the “2026 Three-Body Day: Dehydration Soak Carnival” held in Beijing, Wumeng Xinquan and the Three-Body Universe announced a strategic partnership and unveiled their first joint product, “Zhi Zi Agent 1.0.” This AI conversational agent takes Zhi Zi, the small-universe administrator from “The Three-Body Problem,” as its character anchor and supports multi-turn interactions and cross-session memory.
In the division of labor, Three-Body Universe is responsible for world-building, character boundaries, and personality traits, while Wuwen Xinqiong provides infrastructure such as agent development, knowledge and memory management, and reasoning services. The product is built on the AgentWorks platform, organizing world‑building materials into a searchable knowledge base, managing character tone through hierarchical prompting, and leveraging Infini Memory to maintain continuity of memory across different conversations.
According to Wuwen Xinqiong, during dialogue generation, character expression and world‑view knowledge retrieval proceed in parallel; the initial draft is then reviewed by an independent proofreading module to identify deviations from character design and breaches of established settings, after which any specific conflicts are sent back for revision. The system handles pinpointing clear factual inaccuracies, while whether the character’s expression aligns with the original work is vetted by the Three-Body Universe content team.
The two parties will also explore long-term character memory and relationship evolution, multi-character shared-world narrative, and multimodal interaction.
Tencent Hunyuan has open-sourced Hy-Nomas, which can generate video worlds segment by segment using both text and camera footage.
Tencent’s Hunyuan Visual Generation team has open-sourced the interactive video world model Hy-Nomas and provided an online demo. Users first describe a scene, then use multi-turn text or camera trajectories to control subsequent segments, gradually modifying objects, styles, environments, and character movements.
The model generates text in a autoregressive manner, chunk by chunk, while maintaining the KV cache throughout the generation process to ensure that subsequent frames seamlessly continue the preceding scene. The interactive interface allows users to input additional instructions between chunks and also lets them adjust the camera view using the keyboard. A hands-on test conducted by the WeChat official account “Product Notes” demonstrated first generating a cat in the kitchen, then specifying further actions such as the cat standing up, stretching, and walking forward.
The official repository has made training, inference, demonstration, and data‑annotation code publicly available, with the model weights now hosted on Hugging Face. Developers can fine-tune LoRA models using their own videos; a low‑VRAM configuration supports a single RTX 4090 GPU, and memory requirements are reduced through INT8 weights and caching, sliding‑window history, and CPU offloading—though offloading also increases data‑transfer overhead. The project is licensed under Apache 2.0.
Ji Meng has launched a short‑video studio, enabling scripts, storyboards, and localized reshoots to be completed on the same platform.
On October 26, JIMM AI announced the launch of its “Short Drama Studio,” integrating scriptwriting, art design, storyboarding, post-production, and finished‑product remastering into a single creative workflow. The web platform already offers interfaces for uploading scripts, creating scripts, and initiating remastering and re‑rendering.
According to the official introduction, the in-house script model DreamScript supports genres such as original works, adaptations, international expansion, and commentary, and can process long texts of up to 100,000 characters. The image model Anycook is designed for character and scene design in short-form dramas, while the storyboard model DreamShot supports canvas layouts, storyboards, and 3D director’s stage rehearsals.
The post‑production tools can regenerate individual segments that don’t meet expectations, separate vocals from background audio, or split shots. Remake and re‑rendering can identify the characters, settings, and shot structure of existing short films, then generate different versions tailored to the target style, aspect ratio, and market. The official page also offers intelligent pre‑views, video breakdowns, and upscaling capabilities of up to 4K.
Tencent has open-sourced the Chinese–English simultaneous translation model Hy-SimulMT, which supports real-time listening and translation as well as voice tone cloning.
On October 26, Tencent’s Hunyuan released and open-sourced the Hy-SimulMT simultaneous translation model, which supports bidirectional speech translation between Chinese and English. The model can continuously process the speaker’s audio, generating translations and synthesized speech in real time while listening, and it preserves the speaker’s vocal characteristics through real-time voice cloning. The Tencent Hy Translation app already provides a way to experience the model.
The open-source project includes the SimulInfer inference service and the SimulTrain training tool. The model offers two output modes: translated text interleaved with the original, and a combined version of both. It supports formats such as FP8 and GGUF. When deploying, developers must also configure a TTS service for speech synthesis and voice cloning.
In Tencent’s RealSI long-audio evaluation, the model’s StreamLAAL metric is approximately 3 seconds, which reflects the streaming latency of the translation relative to the input audio, rather than the time taken for the first speech output. According to the GEMBA translation quality score in this evaluation, its performance outperformed both the open-source models used for comparison and four out of five commercial APIs.
The team also conducted a small-scale comparison using 18 segments of approximately six-minute-long audio in both Chinese and English, alongside six professional simultaneous interpreters; however, the results apply only to the specific sample, participants, and test setup used, and cannot be generalized to conclude that this approach can replace human simultaneous interpretation in all scenarios.
Lao Gan Ma denies “reducing staff from 600 to 48,” confirming that AI-based quality inspection has been implemented.
Lao Gan Ma has responded to rumors that “after the AI quality-control system was launched, the number of chili-picking workers was reduced from over 600 to 48.” According to a report by Phoenix News’ “Storm Eye” on October 26, the outlet’s reporter called Lao Gan Ma’s human resources department, where an employee explicitly stated that the online claims about staff adjustments were false.
The staff member confirmed that AI-based quality inspection technology has been deployed at the production base, but the company has never made any personnel adjustments on that scale, adding, “We’ve never employed so many workers before.”
ROG × “Shadow Blade Zero” first batch of co-branded hardware is now on sale, with martial-arts-inspired designs spanning both gaming consoles and peripherals.
The first batch of co-branded hardware for “Shadow Blade Zero,” launched in collaboration between ROG and Lingyoufang S-GAME, went on sale at 10:00 AM on October 26, covering mini PCs, monitors, keyboards, headsets, mice, and keycap sets.
The official product page incorporates a black-and-red color scheme, blade-like textures, weapon engravings, character silhouettes, and crimson bloodlines into the hardware’s design. The NUC 16 collaboration mini PC features a translucent front panel printed with the silhouette of the protagonist “Hun,” and its stand is detachable, supporting both horizontal and vertical orientations.
The NUC 16 Collaboration Edition can be configured with an Intel Core Ultra 9 290HX processor and an RTX 5080 laptop GPU, featuring a triple-fan design and an integrated vapor chamber, and supports tool-free upgrades to memory and the SSD. According to Youmin Xingkong, the co-branded chassis will go on sale separately on October 29; customers who purchase a designated product and complete registration on the official website can participate in a limited‑time giveaway of game redemption codes.
Dianping’s internal beta of “Life Expert”: You can earn growth points by finding stores, checking in, and interacting.
Dianping is currently piloting the “Life Expert” user growth system in select cities. According to multiple users and confirmed by the reporter’s testing, as of October 26, some users in Hefei, Zhengzhou, Fuzhou, and other locations can already see the entry point on the app’s “My” page.
Compared with the previous Lv system, which relied primarily on awarding review points, “Life Home” covers a broader range of everyday activities. Users can earn growth points for reviews, notes, check-ins, public ratings, location tagging, and contribution updates; interactions such as finding stores, liking posts, and replying to others’ reviews are also factored in. The program places particular emphasis on tracking one’s journey and helping other users discover great places.
The new system comprises 18 tiers, ranging from Level 1 to Level 18. Benefits include free trials and in-store gifts, as well as ride-hailing coupons and coupon packages. Long-time users who consistently maintain review and record-keeping habits typically attain higher tiers in the new system. This rollout is currently a gray-scale test in select cities and has not yet been fully rolled out to all users.
On its fourth day of release, “Sanxingdui: A Future Past” grossed approximately RMB 1.19 million, while Bona Film Group’s stock hit the daily limit down.
Following its October 23 release, the theatrical feature film “Sanxingdui: Stories of the Future,” which was created with deep AI involvement, posted a rather subdued box-office performance during its opening weekend. According to data from Maoyan Professional Edition cited by the Daily Economic News, as of noon on October 26, the 100-minute film had accumulated a total box office of RMB 1.192 million, with average attendance of approximately two viewers per screening on both October 24 and 25.
On the afternoon of the 26th, the page showed that there were 8,469 screenings that day, with an occupancy rate of approximately 0.1%. On the same day, the producer Bona Film Group’s stock hit the daily limit down, closing down 10.06%. Previously, the company had reached its daily upper limit on three trading days—September 28, October 9, and October 12.
Bona Film Group has noted in its announcement on October 12 regarding abnormal stock trading volatility that the film’s box-office revenue remains uncertain. The company’s other AI-powered theatrical release, “Unplanned: Operation Guanghan,” has entered the production phase; as of the date of this announcement, it has neither obtained a public screening license nor been officially scheduled for release.
The 2026 “Harmony Starlight Gala” is scheduled for December 5.
The 2026 “Harmony Starlight Gala” will be held on December 5 at the “Spring Cocoon” Stadium in Shenzhen Bay Sports Center. On October 26, CCTV News released an official poster announcing that this event, which blends technology with the arts, will take place once again.
CCTV Video simultaneously released a teaser for the gala, linking last year’s theme, “Together, We Can,” with this year’s, “Together, Toward the Future.” The organizers have positioned this year’s event as another convergence of technology and the arts, aiming to showcase, through the gala, the joint participation of creators and users within the HarmonyOS ecosystem, as well as the evolution of application experiences.
Lu Yang’s produced film “Probe” is set for release on November 21, with a new trailer unveiled.
The film “Probe” has announced its nationwide release on November 21 and unveiled the “Opening the Door to Mystery” trailer on October 26. The movie is produced by Lu Yang and Wang Hongwei, directed by Hou Xiaoyao, and stars Feng Shaofeng, Chen Yusi, and Li Meng in the leading roles. Guo Yue, Che Xiao, Zhou Yemang, and Yang Yi also appear in the film.
The trailer sets the scene in a villa: Ding Boyang, Ma Ya, and Wan Ying engage in a covert power struggle, with suspicion and danger escalating continuously, while someone else lurks in the shadows, spying on them.
#Welcome to follow the iFanr official WeChat account: iFanr (WeChat ID: ifanr), where we bring you the latest and most exciting content第一时间.