TMTPost reports that Qwen Office 2.0, released November 9, adds collaborative spaces, automatic tasks, and direct document editing for team workflows. Its hands-on test found that Qwen4-Max Preview improved completion speed and reduced rework, but WorkBuddy still delivered stronger organization, detail, and overall output quality. Enterprise Standard costs RMB198 per seat per month, while Flagship costs RMB298 per seat per month and requires at least 50 seats.
Intel publicly displayed two samples of its Diamond Rapids "Xeon 7" server processor at the Tencent Global Digital Ecosystem Conference in Shenzhen, cnBeta reported. Intel says the data-center CPU can provide up to 256 Panther Cove performance cores, 1.28 GB of LLC, 16 memory channels and 128 PCIe 6.0 lanes, with compute tiles made using Intel 18A-P. Wccftech reported a planned 2027 launch and a joint-development program with Tencent Cloud CVM.
Google is donating gVisor, including its name and trademarks, to the Cloud Native Computing Foundation (CNCF), shifting the project to maintainer-led governance. Modal, Ant Group and Tines are joining the maintainers, while OpenAI, Tencent and NVIDIA continue contributing. Modal says gVisor powers more than 40 million containers daily and reports that its RDMA implementation reaches about 99% of native bus bandwidth in NCCL benchmarks.
WeChat's HarmonyOS app began an 8.0.23.37 beta invitation in Huawei AppGallery, with testing scheduled from November 10 to December 10, 2026. According to user and blogger feedback cited by IT之家, the build adds AI summaries for official-account articles and introduces Xiaowei features in articles, chats, images, video and other areas. Availability varies because many functions are being gradually rolled out and some require account gray rollout or a minimum app version. IT之家 also reported that the HarmonyOS app had recorded 90.51 million installations.
Tencent launched Yuanbao Search on November 11, 2026, for general users through a standalone website and an engine option in the latest version of QQ Browser. According to Tencent, the service uses Yuanbao's large-model capabilities across query understanding, retrieval, ranking and answer organization, with access to public webpages, WeChat Official Accounts, WeChat Channels, academic literature and specialized channels. It offers AI summaries, follow-up questions, source tracing, academic content tools and labeled advertising that users can collapse. In a hands-on test, Zhidx found that retrieval quality varied by content type: an Official Accounts result included information published that day, while the first page of video results for the same query primarily showed content from a year earlier. The report said further observation was needed to assess whether these differences would improve.
An analysis by 万点研究, also published by 钛媒体 with authorization, compares four AI short-drama strategies: ByteDance’s production-to-distribution loop, Tencent’s WeChat-based viewing-time incentives, Kuaishou’s monetization of Kling AI, and Baidu’s AIGC infrastructure. It argues that platforms are competing not only for scripts and traffic but for control of settlement coefficients that shape content supply, distribution, and revenue sharing. The analysis cites QuestMobile figures showing that Red果 Free Short Drama exceeded 430 million monthly active users in September 2026. Several company metrics and strategic interpretations are cited from QuestMobile or company disclosures and were not independently verified in the supplied reports.
Tencent began the internal beta of 鹅次元, a Windows AI game companion that can reportedly interpret on-screen game information and interact with players in real time. According to Tencent product details cited by DataEye, it uses visual recognition, multimodal dialogue, character memory, text and voice messages, and real-time calls, while supporting companion chat through PC and mobile QQ. The beta includes characters 夏夏, 沈露白, 谷雨, and 贝拉, although DataEye says 贝拉 is not yet open; desktop-pet, wallpaper, speech-substitute, and voice-changing features are described as forthcoming.
A New Berry Daybreak investigation, authorized for republication by 36Kr AI and Cyzone Tech, examines ByteDance’s enterprise push for Doubao Work, including its Feishu integration, seat-plus-usage pricing, sales reorganization and announced opening to third-party models including Kimi 3.1 Pro, GLM5.3 and DeepSeek V4.1 Flash. User examples indicate that Doubao Work can save time on approvals, analysis and reporting, but may omit conditional steps or use outdated information. WorkBuddy, Qwen Office and Baidu Dazi are pursuing enterprise markets through different products and industries. ByteDance has not disclosed Doubao Work’s sales or contract figures.
Rest of World reported that MiniMax has about 16 San Francisco job openings, including a research lead for its core AI model and a recruiter focused on Silicon Valley and Seattle. Moonshot AI is also recruiting U.S.-based solutions architects to help enterprise customers integrate its models. Experts cited by Rest of World said Chinese firms lag U.S. competitors in funding and compensation, limiting their ability to attract researchers from OpenAI and Anthropic.
A commentary from the WeChat account 人人都是产品经理, credited to 怪哥, says researchers born in the 1990s and 2000s are leading important AI work at major Chinese technology companies. It cites Yao Shunyu at Tencent’s Hunyuan, Sun Tianxiang at Baidu, Luo Fuli at Xiaomi’s MiMo, Zeng Yan at ByteDance’s Seedance 2.0, Zeng Aohan at Zhipu AI’s GLM pretraining and infrastructure, and Zhao Pengyu at MiniMax. The commentary argues that scarce AI talent and rapid model development are weakening traditional seniority-based promotion rules. The personnel details and interpretation are not independently corroborated in the supplied material.
WeCom's native AI assistant Da Yuan is reportedly fully available to enterprise users and integrated with Tencent Hunyuan Hy4. It can summarize chats, send messages, read and write documents and spreadsheets, compile project progress, run scheduled tasks, and coordinate with Agent bots such as WorkBuddy inside WeCom.
Tencent released the Hunyuan Hy4 model on November 4 and said its weights would be open-sourced soon, while also introducing the lightweight Hy4 Flash. GeekPark reported that Hy4 ranked among the leading open models based on leaderboard results and demonstrated long-horizon agent workflows, visual understanding, visual coding, and computer operation in hands-on tests. Tencent’s claims about Hy4’s position among the strongest open models and its financial benchmark results remain vendor-supported claims. GeekPark also noted that a gap persists with leading closed models on difficult reasoning and engineering tasks.
A Tencent-provided article says its WorkBuddy office agent has handled more than 9 billion task requests, with token consumption rising over 50-fold after more than 100 version iterations. It cites August data from a third-party product ranking showing WorkBuddy first among Chinese desktop AI office agents by monthly active users, and describes deployments at Mengniu and Shanghai Jahwa. The article also outlines permission controls, sandboxing and execution tracing as governance measures for agents operating in enterprise workflows.
A WeChat article argues that AI agents could gradually take on routine information gathering, status tracking and report preparation, challenging management roles based largely on controlling information flows. It maintains that judgment, coordination, conflict resolution, employee development and accountability still require responsible people. The article cites limited organizational changes at Tencent Games and Baidu, while noting that these changes cannot all be attributed to AI and do not amount to the elimination of middle management. It also presents Gartner's organizational-flattening forecast as a prediction rather than an achieved result. To test the idea, it proposes a 90-day project pilot that measures information handoffs, approval delays, omissions, errors, delivery and employee and customer experience.
Tencent WorkBuddy apologized for failing to promptly announce the end of a 50% credit discount for DeepSeek-V4.1 Flash. The discount launched on September 10, was extended through November 5, and pricing returned to normal at 14:00 that day. WorkBuddy said affected usage from November 5 at 14:00 to November 8 at 23:59 will be repriced at 50%, with the difference refunded automatically within seven working days.
Tencent chief AI scientist Yao Shunyu warned that AI models will soon reach a threshold where their biological capabilities can cause harm, with biological risks potentially harder to manage than cyber risks. Speaking after Tencent launched Hy4, he urged companies releasing open-weight models to take safety seriously, while Tencent said Hy4 excluded software-exploitation material from training and is weaker than leading models at designing experiments based on internal reviews.
A Juejin AI article reviews DeepSeek's October 17 release of V4.1-Flash-1017, highlighting reported gains in Terminal-Bench, Automation-Bench and ProgramBench, plus vendor claims of lower task costs through token and agent-workflow optimization. The author reproduced DeepSeek's “wish-driven programming” racing-game example through DeepSeek Harness, finding that a one-line request produced a playable game with basic driving, tracks and timers, while a more detailed prompt improved its visuals and polish. The reported second run took 51 minutes and 14 seconds, consumed 60.1 million tokens alongside another smaller task and cost about 3 yuan.
Tencent chief AI scientist Yao Shunyu said at a 2026 Future Science Prize dialogue that durable AI differentiation may come from products that understand user backgrounds, use proprietary data and context, and earn user trust. Using WeChat as an example, he argued that access to a user's schedule, relationships and activities may matter more than having the strongest underlying model, though the report does not say such a WeChat AI feature has launched.
WIRED AI reports that Niantic, the creator of Pokémon Go, has pivoted toward world models, with Niantic Spatial CTO Brian McClendon saying the company aims to reconstruct the real world from its data. World models use visual inputs to generate simulated environments that can support robot training and autonomous-driving development, while Wayve uses dashcam footage to recreate hazardous roads. WIRED also notes that training data may create more work for people who record everyday physical activity.
A hands-on report documents an 18-step WorkBuddy workflow that used three natural-language prompts to generate four 3D storyboard images and a 13.7 MB vertical MP4 with Chinese voiceover and subtitles. The roughly 14-minute process included installing edge-tts, inspecting frames, and repairing a subtitle-rendering defect. The author reported that the final MP4 did not play smoothly in the local player, leaving the complete audio, subtitle, and motion output unverified.
Tencent published the Apache 2.0-licensed Qwen3.8-27B-DSpine-b16 checkpoint on Hugging Face, a five-layer, 2B-parameter draft model with a block size of 16 and an 8,192-token sliding window. Its model card says DSpine generates draft tokens in parallel and reports higher acceptance lengths and throughput than DFlash2, with end-to-end results reaching 3.84× autoregressive decoding across the listed agent benchmarks.
In a commentary published by the WeChat account Ning Mingliang Seerning, AI compensation is described as a “bottleneck premium”: buyers may bid according to the potential cost of losing scarce talent rather than pricing compensation strictly by individual output. The report links this mechanism to buyers’ cash-flow and financing constraints, noting that four major overseas cloud vendors have indicated more than $700 billion in combined capital spending for data centers and AI chips this year. It also highlights that prominent compensation figures circulating through recruiter accounts and social media have not been confirmed by the relevant companies.
DoNews reports that China’s mini-game revenue reached RMB31.657 billion in the first half of 2026, up more than 35%, while in-app-purchase revenue rose 38.12% year over year. User-acquisition costs increased by more than 400% in some categories and average creative lifespans fell below five days. According to DataEye data cited by DoNews, AI-generated materials accounted for 46% of mini-game advertising creatives, nearly matching conventional video at 48%.
DataEye Research Institute monitored 51,800 new AI dramas and animated dramas on Douyin’s native platform in September, with 372 titles exceeding 100 million views. In the monitored sample, 2,844 titles with more than 10 million additional views accounted for 86.9% of new-title viewing growth. The report attributes the supply surge partly to lower production barriers and argues that platforms have commercial advantages through audience access, advertising, distribution and user data. Play counts and aggregate platform financials do not reveal individual creators’ earnings or establish how revenue is distributed.
The 2026 Singularity Intelligence Technology Conference is scheduled for November 20–21 at the Beijing Wanda Wenhua Hotel. According to QbitAI, its agenda covers AI capabilities, agents, AI-native software development, world models, Physical AI, infrastructure, and enterprise applications, with a concurrent C++ and systems software conference.
Tencent unveiled Hunyuan Hy4 at the Wuzhen Summit, adding visual understanding and computer-operation capabilities. According to the report, the model can interpret charts and documents and operate graphical interfaces and command-line tools. The report inconsistently describes the release as fully open source while also saying the model weights will be released soon.
China’s State Council General Office issued an 18-task opinion calling for cautious AI deployment in labor-intensive industries, employment-risk assessments for projects that could cause substantial job disruption, and priority use in labor-scarce or hazardous jobs. At the Wuzhen Summit on November 4, the China Academy of Information and Communications Technology joined 20 organizations, including Huawei, Tencent, Ant Group, Alibaba and ByteDance, in issuing an eight-point AI-agent self-regulation convention covering interoperability, user awareness and choice, final user control, and opposition to algorithmic exploitation. The commentary, published by the WeChat account Qiongqi Business and written by Qiongqi Observer, interprets the two documents as a map for the pace and boundaries of AI deployment. That investment interpretation is the author’s analysis, not a binding implementation rule.
TMTPost reported that DeepSeek completed a second financing round of RMB 105.5 billion at a post-money valuation of RMB 615.5 billion, with nearly 30 institutions participating. The report said the newly established Hengdian Baihui fund, managed by Hengdian Capital, has a total size of RMB 1.65 billion and received about RMB 1.5 billion of the financing. Its 16 limited partners are mostly Zhejiang-based industrial business owners, with the Hengdian group accounting for about 33% of commitments.
A firsthand beta review by Lei Technology, republished by 36Kr with authorization, found that Tencent’s Chatterfly can turn speech into polished text and apply Skills for prompts, project tracking, marketing copy, work reports, and meeting minutes. The reviewer also tested web Q&A, follow-up questions, editing selected text, translation, global invocation, and terminology correction. Chatterfly remained invitation-only on macOS and Windows, while iOS and Android versions had not launched.
Tencent's Lightweight Cloud team open-sourced LoopForge, a cloud-based AI development platform built around persistent workspaces, shared multi-role agent sessions, gated SOP pipelines, unattended issue automation, and engineering-level recursive self-improvement. InfoQ reported that the platform supports tools including Claude Code, Codex, and CodeBuddy, while also drawing on DeepSeek's Super-agent research. The team says LoopForge is used internally across more than 20 businesses and automatically completes over 200 requirements and routine work items per month.
What problem it solves
AI agents need isolated environments that can execute code and workloads while supporting fast creation, secure network access, and manageable state. They also need an interface that can integrate with existing agent tooling and deployment workflows.
Key features
Hardware-level isolation — Each sandbox runs in its own MicroVM with a dedicated operating system kernel. Kernel-…
What problem it solves
Real-world translation involves more than converting text between languages: tasks can require terminology, style, personalization, delimiter preservation, and structured-data constraints. Hy-MT2 is designed to support translation among 33 languages and follow translation instructions in multiple languages for complex real-world scenarios.
Key features
Instruction-driven tas…
What problem it solves
A coding-agent run can skip planning, omit tests, review its own work, or stop before completion, making delivery inconsistent and difficult to supervise. Teams need repeatable workflows that separate roles, execute work in sandboxes, preserve review evidence, and require human approval before changes ship.
Key features
Repeatable SOP delivery — Standard operating procedures…
What problem it solves
Teams need to bring documents together so they can be searched, reasoned over and kept up to date. They also need a way to move from source-grounded retrieval to multi-step tasks while maintaining organized, current knowledge.
Key features
Three knowledge workflows — RAG supports lookup, an agent handles multi-step tasks, and a wiki organizes knowledge. All three operate on …