Ji Meng has officially launched its “Short Drama Studio”; HTC has unveiled the VIVEE pocket AI interactive companion; and OpenAI has released its new low-cost model, GPT-6.1 Terra.
Anthropic has officially embarked on its path to an IPO. The company, founded just five years ago and at the heart of the generative AI boom, is poised to set a record with this offering.
Anthropic disclosed in a filing with the U.S. Securities and Exchange Commission (SEC) that it plans to list on the Nasdaq Stock Market under the ticker symbol “ANT,” with an IPO targeted for the week of November 16. The company will issue Class T special shares, which will be held by Class T directors.
In its prospectus, the company disclosed that first-half revenue totaled $16.29 billion, compared with $1.16 billion in the same period last year; net loss amounted to $80.75 billion, versus $12.97 billion a year earlier; and operating loss stood at $693 million, down from $2.14 billion in the prior-year period.
During the nine months ended September 30, no single customer accounted for more than 10% of the company’s revenue. Seventy-nine percent of revenue came from customers other than the top five. In terms of collaboration, under its agreement with SpaceX AI, the company has agreed to incur cumulative expenditures of up to USD 120.5 billion over a period ending in 2029.
The company disclosed that capital expenditures will increase significantly and that it may need to seek additional financing sooner than currently anticipated; it will not pay any dividends in the foreseeable future and has not yet entered into any agreements for major acquisitions; it may issue convertible notes with a total principal amount of up to $42 billion.
During the nine months ended September 30, no single customer accounted for more than 10% of the company’s revenue. Seventy-nine percent of revenue came from customers other than the top five. In terms of collaboration, under its agreement with SpaceX AI, the company has agreed to incur cumulative expenditures of up to US$120.5 billion over a period ending in 2029. (Source: Sina Tech)
Shortly after the iPhone Duo’s launch, some users reported hinge malfunctions that caused cracks in the inner display. Apple has now responded to these reports in a statement to MacRumors, stating that only a small number of devices are affected.
Apple did not explain the cause of the cracks, nor did it disclose whether a repair solution would be offered. Affected users can visit an Apple Store to exchange their devices for new ones, but due to limited stock, some may have to wait.
Apple stated that the issue affects a very small number of users. The vast majority of users will not experience iPhone Duo screen malfunctions, nor need they worry that the problem will gradually emerge with continued use. For devices that do develop faults, cracks typically become apparent shortly after purchase.
According to reports, some of the crack images circulating online were AI-generated, further complicating efforts to assess the true extent of the issue.
Many of the problematic iPhone Duo units have cracks concentrated in the lower half of the inner display, running along the hinge. Users report that before the screen breaks, they experience a jerky, laggy feel, and sometimes even hear a clicking sound. The repair‑and‑disassembly site iFixit disassembled the iPhone Duo and identified several potential causes of hinge failure. According to the site, the hinge area houses multiple components; in particular, a central support rod inside the hinge that should retract may fail to do so properly, leading to screen damage. (Source: IT Home)
OpenAI announced early this morning the launch of its new model, GPT-6.1 Terra, which focuses on completing complex computer tasks at a lower cost, particularly excelling in tasks that require visual reasoning and manipulation of desktop software. Compared to GPT-5.6 Terra, the new model also features significant improvements in factual accuracy and programming capabilities.
Terra is primarily aimed at enterprises looking to reduce automation costs and can be used for tasks such as processing invoices and interpreting technical drawings. According to OpenAI, even when tackling highly challenging professional tasks, Terra can deliver results at a cost far lower than that of high-priced models. In the Agents’ Last Exam benchmark test, which evaluates complex professional work, Terra’s performance improved significantly. The test items included creating 3D models from architectural blueprints and meticulously cataloging film shots for use by film restoration teams.
When using medium reasoning intensity, Terra’s score is 15 percentage points higher than GPT‑5.6 Terra, while its cost to complete tasks is only about one-third of the latter’s. Compared with GPT‑6 Astra, Terra achieves comparable test results, yet its price per token is only about one-tenth of the latter’s.
Terra also has an advantage in speed. OpenAI claims its task execution speed is more than three times that of GPT-6 Astra and GPT-6.1 Sol. At the highest reasoning intensity, the cost per task is only about one-ninth of Claude Opus 5.5.
GPT-6.1 Terra was made available on October 27 to Plus, Pro, Enterprise, and Edu users, and can be used in ChatGPT Work and Codex. Developers can also invoke it via the OpenAI API.
In terms of standard API pricing, each 1 million input tokens costs $1 (approximately RMB 6.7 at the current exchange rate), each 1 million cached input tokens costs $0.05 (approximately RMB 0.33 at the current exchange rate), and each 1 million output tokens costs $5 (approximately RMB 33.4 at the current exchange rate). OpenAI states that the new model’s pricing is 50% lower than GPT‑5.6 Terra and GPT‑6.1 Sol. According to OpenAI’s preparedness framework assessment, Terra’s cybersecurity capabilities have reached the “critical” threshold. (Source: IT Home)
On October 26, Meta announced a major financing plan totaling up to $95 billion (note: at the current exchange rate, this is approximately RMB 634.357 billion) to provide funding for the continued development of its core business and emerging artificial intelligence initiatives.
This $95 billion is divided into several components, including a $22.5 billion forward sale of Class A common stock, $12.5 billion in mandatory convertible preferred shares, and up to an additional $60 billion in potential Class C common stock issued at market price. NVIDIA, AMD, and Broadcom all intend to purchase up to $5 billion worth of Meta’s Class A common stock.
Meta stated in a press release that its investments in artificial intelligence have yielded tangible returns by boosting user engagement, enhancing ad performance, and increasing profitability.
The company expects capital expenditures to reach US$135–145 billion by 2026, with Meta planning to finance its AI investments through a balanced mix of operating cash flow, investment-grade debt, strategic partnerships, and equity capital. (Source: Sina Tech)
According to QuestMobile data, as of August this year, Tencent’s “Lü Bao Pao” short-drama mini-program has surpassed 100 million monthly active users, reaching 143 million. In the first week of October, its weekly active users totaled 89.68 million, a 44.1% increase compared with the first week of September. Since its official rebranding in July, the product’s weekly active user base has neared 90 million within roughly three months.
DataEye monitoring data shows that from October 1 to 7, Lvbubble short-form dramas collectively deployed 43,000 ad creatives, with the daily average volume up 57% compared to the week before the holiday, as both content supply and advertising investment were ramped up in tandem.
From a growth‑driving perspective, the full replacement of the original “Short Drama” mini‑program entry, coupled with traffic support from the WeChat ecosystem, has delivered direct incremental growth for Green Bubble. Leveraging WeChat’s distribution network, Video Accounts—boasting 600 million daily active users—plays a pivotal role; meanwhile, Green Bubble continues to expand its content offerings, covering live‑action series, AI‑generated dramas, anime series, and more.
“Green Bubble Short Dramas” is a WeChat mini program developed by Tencent. The app’s interface is divided into three sections: “Free Theater,” “Swipe Through,” and “My.” Under “Free Theater” are categories such as “AI Dramas,” “Live-Action Dramas,” “New Releases Daily,” “Anime Dramas,” “Must-Watch This Week,” and “Million-View Hits.” Tapping the “Swipe Through” entry allows users to scroll up and down, accessing an immersive recommendation feed. All episodes are free, though ads are interspersed between them. (Source: IT Home)
On October 26, Tencent’s Hy-SimulMT large-scale simultaneous translation model was released and open-sourced today. It features low latency and high accuracy, supporting real-time speech-to-speech translation between Chinese and English. The speaker’s audio can be fed into the Hy-SimulMT model in real time, which transcribes and translates on the fly, delivering both text and speech output instantly. It also supports real-time voice‑style cloning, enabling highly faithful reproduction of the speaker’s voice and tone. The model made its debut in Tencent’s Hy Translate app, offering a free trial experience.
Low latency is a major highlight of the model; the Hy-SimulMT large-scale simultaneous interpretation model has been measured to deliver response times of around 3 seconds, while the average human professional interpreter takes 4–5 seconds. Evaluation results show that this model’s translation performance surpasses mainstream open-source simultaneous interpretation models on the market and even outperforms most commercial simultaneous interpretation APIs.
According to the official statement, on RealSI—the widely used open-source evaluation dataset for long-form audio in the industry—Hy-SimulMT has demonstrated strong performance in simultaneous interpreting. To more accurately assess Hy-SimulMT’s true capabilities, the team, within the bounds of compliance, extensively collected newly available Chinese and English audio from the internet, covering ten domains: software AI, medicine, manufacturing & energy, natural sciences, agriculture & food, business management, law, education, culture & tourism, and sports & lifestyle, and encompassing various scenarios such as speeches, interviews, and conferences. On this dataset, the team not only evaluated Hy-SimulMT’s performance but also invited multiple human simultaneous interpreters to provide their own interpretations of these audio recordings.
There are a total of six human simultaneous interpretation experts, three of whom hold CATTI Level 1 interpreting certificates and three hold CATTI Level 2 interpreting certificates; all have many years of hands-on experience in simultaneous interpreting. The evaluation compared the translation output of Hy-SimulMT with that of the human experts in terms of latency and translation quality. Among the human experts, the highest-scoring candidate from among the three was selected. In this capability‑assessment test, Hy-SimulMT achieved translation quality on par with the human experts (even scoring higher) while delivering faster performance (lower latency). (Source: IT Home)
On October 26, it was announced that JIMENG AI has launched a new feature called “Short Drama Studio.” As a one-stop AI tool designed for short‑drama creation, “Short Drama Studio” integrates the entire workflow—script organization, character and scene management, storyboard generation, video production, and post‑production editing—into a single platform, eliminating the need for creators to switch between multiple applications.
In March this year, Xiaoyunque, a content-creation platform under CapCut alongside Jiumeng, was the first to launch Short Drama Agent. Reportedly, certain features of the “Short Drama Studio” have been integrated with the same capabilities as Xiaoyunque’s Short Drama Agent, and have been further updated and upgraded on that basis.
JiMeng and Xiao Yunque have each previously explored different tool formats and built strong reputations, with JiMeng particularly leading the way in adapting its capabilities to the eedance series of models. This integration of capabilities allows creators to find a wider array of tools on a single platform, enabling them to better leverage their respective strengths.
A representative from Jiameng stated that while the forms of content creation are constantly evolving, creators universally seek efficient, unified创作 tools. Jiameng aims to continuously integrate proven AI‑powered creative capabilities into a single platform, enabling creators of all types to complete the entire workflow—from inspiration to finished output—within familiar tools, allowing them to focus more on the creative process itself. (Source: Sina Tech)
On October 26, according to a report by Science and Technology Daily citing the National Key Laboratory of Optical Communication Technologies and Networks at China Information and Communication Technology Group, the laboratory recently collaborated with Pengcheng Laboratory and FiberHome Fujikura Optical Fiber Technology Co., Ltd. to successfully complete a real-time bidirectional transmission test over a multi-core fiber at 3.2 Pb/s (petabits per second), setting a new world record for single-fiber bidirectional real-time transmission capacity.
“A transmission rate of 3.2 Pb/s means that more than 20,000 4K movies could be downloaded in a single second,” said Yang Chao, a senior optical systems engineer at the National Key Laboratory for Optical Communication Technologies and Networks of China Information and Communication Technology Group. He added that this solution has validated the industrialization potential of the multi-core fiber ultra-broadband capacity‑expansion roadmap, providing next‑generation high‑speed optical communication core technology to support interconnection among large-scale computing hubs, backbone transmission, and 6G transport.
With the rapid advancement of artificial intelligence, large‑model computing, and the scaling up of data centers, network traffic is experiencing explosive growth, pushing traditional single‑mode fiber transmission systems ever closer to their capacity limits. To address this bottleneck, our team has innovatively proposed a 24‑core multi‑core fiber bidirectional transmission architecture. By deploying a 10.3‑kilometer 24‑core multi‑core fiber link and integrating S+C+L three‑band multiplexing with space‑division multiplexing, we have established thousands of parallel transmission channels, dramatically surpassing the conventional single‑mode fiber’s bandwidth ceiling. The system is equipped with commercial coherent optical modules and employs high‑order probabilistic shaping modulation to enable stable, real‑time bidirectional signal transmission. Furthermore, optimized core‑layer arrangement effectively suppresses inter‑core crosstalk and simplifies signal equalization, balancing transmission performance with practical engineering considerations.
“Compared with the capacity ceiling of about 100 Tb/s for conventional C+L-band commercial single‑core G.652 optical fiber, this system increases the total per‑fiber transmission capacity by a factor of 32. It has also successfully demonstrated real‑time transmission using a self‑developed optical module—400 Gb/s per wavelength in the S band and 1.2 Tb/s (terabits per second) per wavelength in the C+L band—thereby breaking through the per‑core fiber capacity barrier that is difficult to overcome by spectral‑efficiency scaling alone,” explained Yang Chao. Currently, these results have been submitted as an invited paper to the internationally renowned optics journal *Journal of Lightwave Technology*. (Source: IT Home)
Apple on Monday unveiled a redesigned iPad mini, featuring an OLED display, water resistance, and a faster chip, kicking off its new-product launch week.
This tablet is priced at $699 and features an 8.5-inch OLED display that delivers deeper blacks, brighter whites, and higher contrast. By narrowing the screen bezels, Apple has increased the display size from the previous 8.3 inches to 8.5 inches. The redesigned model also incorporates a new “Acoustic Display” audio system, with speakers positioned beneath the display, making it the first iPad to achieve IP67 dust- and water-resistance ratings.
The new iPad mini is 14% thinner and 13% lighter than the previous generation. Apple has also repositioned the front-facing camera, making it better suited for FaceTime calls and video conferencing.
The new model is powered by the A20 Pro chip, an upgrade over the A17 Pro processor used in the previous generation; the prior model started at $599. The latest version offers both a standard display and a matte nano-texture display option, with storage capacities ranging from 128GB to 512GB. The top-of-the-line configuration, featuring the highest storage capacity, an etched matte screen, and cellular connectivity, retails for $1,249.
The launch of this new iPad mini kicked off Apple’s late-October lineup of new products. Later this week, the company plans to unveil a new iMac equipped with the M6 chip and a 14-inch MacBook Pro, and will introduce its first touchscreen MacBook Pro before the end of the week. (Source: Sina Tech)
On October 26, HTC announced a pocket-sized AI interactive companion called VIVEE, with a pre-sale price of $99.99 (approximately RMB 667.7 at the current exchange rate).
When users stroke, pat, or rotate VIVEE, the device responds with engaging interactions; it also supports chatting and gaming, and through its interactions with users, this pocket companion gradually develops its own unique personality.
HTC also offers VIVEE Suits styling accessories, which not only give VIVEE a fresh look but also unlock different on-screen UI expressions.
On the hardware side, the HTC VIVEE features a 3D form factor measuring 60×60×50 mm and weighing 114 g. It is equipped with an OLED display, offers all-day battery life, and supports Wi‑Fi, Bluetooth, and NFC connectivity. It also includes haptic feedback and IPX4 water resistance. The device comes with physical on/off switches for both the microphone and speakers, and HTC has pledged that users’ personal data will not be used for training AI models. (Source: IT Home)
HachimoDock, an open-source AI hardware device incubated by Xiaomi’s Technology Committee’s Innovative Products Team, has made its debut. This compact desktop AI device can be freely customized to suit individual preferences, integrating features such as a personalized desk companion, voice-controlled PC agent, smart home integration, and the ability to create custom components.
According to reports, the HachimoDock’s self-built hardware costs approximately 150 yuan. The team aims to lower the barrier to entry for AI hardware, making it the first AI device for beginners to explore Vibe Coding and experience the joy of “creating” a versatile desktop companion using natural language.
This device offers far more than just these functions: users can simply voice their ideas to collaborate with AI, instantly generating and running new components on the spot. Thanks to this open architecture, it can serve as a quick voice-input microphone, a mini gaming console preloaded with tens of thousands of games, or a compact FM radio—while also doubling as a secondary desktop display for monitoring AI task progress and system status, or even a central control hub for smart home devices… The possibilities are endless, left for each user to craft themselves.
Users can interact directly with the desktop pet via voice; it translates user intent into various tasks and dispatches them to computer-based agents such as Codex and WorkBuddy—for example, searching for information, organizing files, or writing code. Meanwhile, the HachimoDock screen simultaneously displays task statuses like thinking, executing, and waiting for confirmation, and provides feedback on progress through the desktop pet’s expressions, movements, brief prompts, and audio alerts.
HachimoDock’s voice control also extends to home‑automation scenarios. By integrating with Xiaomi Miloco, users can issue voice commands to manage their smart home devices. The compact display can even switch to a home‑control dashboard, showing device statuses, lighting controls, and live monitoring feeds—turning a single desktop device into both a computer agent and an interactive gateway for smart home systems. Currently, HachimoDock has made its source code publicly available on GitHub, along with a hardware bill of materials, structural‑part models, and assembly instructions, allowing users to customize it according to their own ideas. (Source: IT Home)