Tencent published an Apache 2.0-licensed repository for WeMM-Video-2B-AWQ, an AWQ W4A16 Safetensors checkpoint with 4.86 GiB of weight files. The model card says the 3B-parameter model accepts text, images, audio, and video and generates text responses in Chinese or English, while the quantization uses group-128 INT4 for feed-forward weights and BF16 for activations and attention weights. Tencent reports successful loading and generation checks on one NVIDIA L20 using Transformers 5.2.0 and a patched vLLM 0.18.0.
Tencent published a Hugging Face repository for the NVFP4 W4A16 Safetensors variant of WeMM-Video-2B under the Apache 2.0 license. The checkpoint contains 4.90 GiB of weights and accepts text, images, audio, and video for Chinese or English text responses. Tencent documents deployment with patched vLLM 0.31.0 using ModelOpt Mixed Precision and Marlin.
Tencent’s Hugging Face listing introduces Hy4-base as a 771B-parameter image-text-to-text model built on a 770B-parameter Mixture-of-Experts language backbone, with 49B parameters activated per token and a listed 1M-token context window. The model card says it supports text, image and video understanding, and provides Transformers support, GGUF weights, a finetuning pipeline, and deployment recipes for vLLM and SGLang. Tencent also reports benchmark and application improvements, but the supplied listing is the only source and its performance claims have not been independently verified.
Tencent released WeMM-Video-2B, a dense 2B checkpoint in its bilingual WeMM-Video family, under the Apache-2.0 license. The Hugging Face model card says it accepts text, images, audio, and video, generates text responses, and offers BF16 weights plus FP8, AWQ, GGUF, and NVFP4 variants.
游戏茶馆 compiled 53 publicly disclosed game investment records through November 11, 2026, finding that industry funds, publisher and game-company investments, incubators, and local government funds remained active. The report says investors increasingly favor teams with proven delivery records or playable prototypes, while AI tools, platforms, and interactive experiences are becoming easier to present as investment narratives. It also cautions that a shift toward AI does not mean capital is leaving games, although the prospect of model companies entering game investment remains a forward-looking claim from BBX Ventures founder Yang Wenfeng.
A 36Kr report and commentary by Kuaidao Qingyi connect sharply rising AI token demand with difficulty for model companies to generate profits. OpenRouter data cited by 36Kr show tokens routed through the platform rising about 250-fold to 128.4 trillion over 20 months. In the second week of September, open-weight models handled 72% of those tokens but accounted for only 27% of platform spending. The commentary reports that Salesforce introduced role-based token budgets, Apple reduced some engineering quotas, and Uber imposed per-person, per-tool limits after exhausting its annual budget in four months. It also argues that outcome-based pricing remains difficult for general-purpose AI because business results can be hard to define and attribute among AI, people and other factors.
TMTPost reports that Tencent’s AI-driven capital expenditure rose from RMB52.8 billion in Q2 to RMB65.1 billion in Q3, up 225% year over year, as infrastructure expanded for model upgrades and WorkBuddy inference. Tencent’s operating margin fell from 33% to 29%, while free cash flow swung from negative RMB13.8 billion in Q2 to negative RMB69.1 billion in Q3, though it returned to positive RMB16.9 billion when compute-related prepayments were excluded.
Tencent announced an update to the WorkBuddy desktop app that reorganizes the task sidebar into Pinned, Recent, and Groups sections. Users can create and reorder groups, assign tasks by dragging or menus, pin frequently used items, and associate each group with up to 16 local folders for Agent access; the features are available in WorkBuddy desktop 5.11.3 and later.
Registry records reviewed by 36Kr, Zhidx and IT Home show that Hangzhou Chengyan became a shareholder of DeepSeek operator Hangzhou DeepSeek, whose registered capital rose from about RMB 16.45 million to RMB 19.92 million. Hangzhou Chengyan added 31 partners, including companies linked to Tencent, CATL, China Mobile, China Telecom, NetEase, Midea and Geely, while founder Liang Wenfeng left its direct partner list. Hangzhou Chengce, which Liang controls, remains its executive partner.
InfoQ visited the exhibition area of the 2026 Tencent Global Digital Ecosystem Conference and surveyed AI offerings spanning agent hosting and governance, agent infrastructure, MaaS services, and training and inference acceleration. The report included demonstrations and covered the showcased products’ technical highlights and application scenarios.
Tencent’s WeVisForensics repository releases five LoRA adapters for WeVisForensics-4B, built on Qwen3.5-4B and trained with DTPO. According to the repository, a main investigator agent coordinates inspector sub-agents that examine visual appearance, physical consistency, common sense, on-screen text, speech content, and voice profiles. Deployment options include merged models for multi-GPU use, five adapters served through vLLM on one GPU, and an in-process Transformers demo requiring at least 24 GB of GPU memory.
At Tencent’s 28th-anniversary executive meeting, Ma Huateng said rapidly advancing AI threatened even established core businesses and called the company’s response a survival issue rather than an ordinary adjustment. TMTPost interprets the warning alongside Tencent’s higher AI spending, proposed organizational changes, and ByteDance’s expansion across content, commerce, workplace software, and AI assistants. The analysis also examines Tencent’s Xiaowei assistant in WeChat and ByteDance’s effort to bypass established app ecosystems with agentic automation.
Tencent reported third-quarter revenue of RMB207.34 billion, up 8%, while Non-IFRS operating profit fell 6% to RMB68.07 billion as AI-related spending exceeded its current returns. Capital expenditure rose 527% to RMB81.43 billion, and TMTPost reported a RMB69.1 billion free-cash-flow outflow and a shift from RMB58.2 billion in net cash to RMB13.7 billion in net debt. Tencent’s Hong Kong shares fell about 6% after the results, while Morgan Stanley cut its target price nearly 13% to HK$480 but retained an Overweight rating.
An article from the Silicon Research Institute’s WeChat account, authorized for publication by 36Kr, reports that ByteDance, Tencent, Alibaba, Baidu and Kingsoft Office recently adjusted their AI-office organizations. ByteDance integrated Doubao, Feishu and Volcano Engine; Tencent moved QClaw-related businesses and teams into WorkBuddy; Alibaba combined three products into Qwen Office and assigned it to the DingTalk team; Baidu consolidated internal products into Baidu Dazi; and Kingsoft Office shifted more of its focus to AI. The article argues that as widely available models become interchangeable, data usability, enterprise-process customization and established customer and commercial channels are more important competitive factors.
A poster argues that human-agent and agent-agent collaboration could make products that rely mainly on person-to-person interaction relatively less important. The poster says WeChat message duration fell from above 80 minutes to above 70 before WeChat Channels and Mini Games helped bring it back above 80, and frames AI agent collaboration as the beginning of a challenge to Tencent, especially WeChat. The message-duration figures and strategic assessment were not independently substantiated in the supplied report.
IT之家 reported that Tencent President Martin Lau said the company’s Yuanbao AI assistant had not performed as successfully as expected and that Tencent had substantially reduced investment in it. Yuanbao is no longer focused solely on user scale, with model capability becoming its priority. Lau also said Tencent was expanding the WeChat Xiaowei test pool, which had reached 30 million daily active users, and expected Xiaowei’s quarterly cost after a full launch to be below Yuanbao’s earlier quarterly cost.
Tencent said chip restrictions will make training its next-generation model take longer. The supplied report does not specify the restrictions, the model, or the expected delay.
Tencent WorkBuddy's desktop version 5.11.3 and later introduces a task-management upgrade that renames “Tasks” to “Recent” and upgrades “Spaces” into “Groups.” Users can organize, sort, and pin tasks and groups, while each group can link to up to 16 local folders with a designated primary folder for WorkBuddy to prioritize; removing a link does not delete local files.
WeChat’s HarmonyOS app released build 8.0.23.46 as an invitation beta through Huawei AppGallery, with testing scheduled from November 18 to December 18, 2026. The official update notes only mention fixing known issues. Bloggers reported Xiaowei AI features and other interface changes across the wider 8.0.23 release, rather than specifically build 8.0.23.46. Many reported features are subject to staged rollout, account eligibility, gray testing, or device and version requirements.
Together AI Blog says Tencent Hy4 is now available through serverless and dedicated endpoints under the model string “tencent/Hy4.” The model supports text and image input, tool calling, JSON output, and a 1M-token context window, with serverless pricing of $0.83 per 1M input tokens, $2.50 per 1M output tokens, and $0.042 per 1M cached input tokens. In Together AI’s tests, Hy4 extracted values from a chart, recreated it in HTML, and generated a playable browser game from a single prompt.
Tencent reported third-quarter revenue of 207.336 billion yuan, up 8% year over year, and attributable net profit of 63.612 billion yuan, up 0.8%. Management said the company’s current capital expenditure is primarily for AI infrastructure that can support continued model training and token generation. Tencent executives also said tighter chip restrictions are reducing available compute, whether leased or self-built, and that training the next-generation model may take longer.
At an analyst meeting, Tencent President Martin Lau said WorkBuddy generated more than $500 million in revenue over the past seven days on an annualized basis, more than twice Tencent Meeting’s figure. He also said third-quarter revenue was more than five times the second-quarter level and that 60% of China’s top 100 enterprises had become customers. IT之家 reported that WorkBuddy’s enterprise edition had reached more than 50 industries and covered 50% of China’s top 500 enterprises. The report did not explain how the seven-day revenue was annualized or how it was compared with Tencent Meeting.
Tencent executives said current capital expenditure is primarily building reusable AI infrastructure for model training and ongoing Token generation, according to IT之家 AI’s report on the earnings call. They expect infrastructure built in 2026 and 2027 to ease operating pressure by 2028, while lower equipment prices and improving training and inference efficiency could improve returns.
Tencent is expanding the WeChat Xiaowei user test pool, which reportedly has reached 30 million daily active users, according to IT之家 citing Tencent President Martin Lau. Lau said Xiaowei's quarterly cost after a full launch would be lower than Tencent Yuanbao's quarterly cost earlier this year. WeChat also said Xiaowei uses the customized WeLM model with low, declining inference costs, while some inference tasks could eventually run on users' devices.
A supplied headline reports that Tencent’s domestic gaming revenue grew 11% in the third quarter. It also describes Tencent’s gaming-AI applications and positioning as continuing to lead the industry, but provides no specific models, deployments, performance results or independent supporting measurements.
TMTPost, republishing an authorized Financial Story Hub article, contrasts Robin Li's public advocacy for AI with Pony Ma's internal warnings about the need to respond quickly. The article says Li described the next phase as the “intelligent organization” at Baidu World Conference on November 12, while leaked minutes from Tencent's 28th-anniversary LinkTime urged staff to split between defending existing businesses and pursuing new initiatives. Its central comparison is interpretive: Baidu is portrayed as an early mover that has not converted that position into user and commercial traction, while Tencent is presented as a later entrant using its financial position and existing ecosystem to catch up.
Tencent’s third-quarter revenue grew 8% year over year, and the company is accelerating its AI transformation. The supplied report consists only of a headline and provides no revenue amount, business-segment details, or specific AI initiatives.
Tencent said WorkBuddy had more than 10 million daily active users and consumed over 100 trillion tokens per day, while revenue from the prior seven days annualized above $500 million. Tencent also said third-quarter WorkBuddy revenue was more than five times its second-quarter level, while AI demand helped cloud-service revenue rise 35% year over year. These adoption, revenue and usage figures came from company disclosures and an earnings call and were not independently verified.
Tencent's open-source LoopForge is described as a self-deployable workbench that organizes existing coding CLIs with roles, Docker workspaces, script checks, structured artifact contracts, and human approval gates. Its workflow engine can reset an upstream target and its downstream nodes for rework, while keeping model conclusions separate from deterministic validation and manual decisions. The Juejin analysis notes deployment and security boundaries, including shared-workspace write conflicts, host Docker socket access, and the need for application-level idempotency.
Tencent disclosed in its 2026 third-quarter financial results that WorkBuddy had more than 10 million daily active users and exceeded 100 trillion Token calls per day as of early November. The company reported that the platform had completed over 100 major updates since March, while its open platform covered 55 professional scenarios and more than 30 vertical sectors. Tencent also said WorkBuddy Enterprise had reached over 50 industries and half of China’s top 500 enterprises.
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 …