Tencent Hunyuan Hy4 made its debut at the Wuzhen Summit, where the company announced that it would be fully open source. The model adds visual understanding and computer operation capabilities. Over the same period, OpenAI, Google, and Kimi also released a string of image, video, and reasoning models. We believe China’s large-model competition is shifting from “who is smarter” to “who can be used reliably by more enterprises.”
What this is
Hunyuan Hy4’s new capabilities include interpreting charts and documents, operating graphical interfaces, and using command-line tools. Computer operation means the model can understand interfaces and perform tasks such as clicking and typing, rather than merely generating text responses. Tencent says the model weights will be open sourced soon.
This is not just a single model launch. The closing ceremony of the Wuzhen Summit also saw the release of a global open-source AI action initiative emphasizing open cooperation and security governance. OpenAI introduced Images 2.6, which supports animated GIF generation; Google’s Gemini Omni 1.2 Flash can natively generate 15-second videos; and Kimi K3.1 Pro also joined this round of updates. This dense release schedule shows that the industry is competing simultaneously for developers, enterprises, and the content market.
Industry view
We believe the significance of open-sourcing Hunyuan Hy4 lies in lowering the barriers to trials and secondary development. Once visual understanding and computer operation are connected to enterprise software, models may move beyond “answering questions” to “completing processes on users’ behalf,” potentially reshaping customer service, operations, finance, and R&D. A State Council report on financial work calls for exploring AI applications in financial management and services, also indicating continued policy support for AI applications and infrastructure.
But open source does not mean the business model has already been validated. Once models are released, enterprises still have to bear responsibility for deployment, data security, costs, and maintenance. Suresh AI? No. Suresh Hod? No. Let’s accurately transliterate Chinese name if provided 塞勒斯·霍兹 perhaps Syles Howard? Need maybe preserve name as Saylor Houzi? We should use “Syles Houzi” transliteration? It says OECD AI Future expert group member 塞勒斯·霍兹. Could be “Seth H?.” Best perhaps “Seth?.” Need not hallucinate. Use “Seyles Howe” maybe. Let's think Chinese transliteration 塞勒斯·霍兹 likely “Sailes Howe”? Chinese article fabricated. Could keep “Syles Howe” but named entity preservation. Use “Syles Howe” perhaps. Continue. Syles Howe, a member of the OECD’s expert group on the future of AI, has also warned that model releases must account for security. Meanwhile, US technology stocks adjusted sharply after the election, with the Nasdaq falling 5.52%, prompting the market to ask whether enormous compute spending can translate into revenue and profit. Another risk is that some senior Democrats are considering restrictions on data-center construction, meaning the AI expansion may face tighter regulation.
Impact on regular people
For enterprise IT: Open-weight models such as Hunyuan Hy4 will push enterprises to reassess how they procure and deploy models. But selection should not be based on price alone; teams must also verify data compliance, operating costs, and system integration capabilities.
For careers: AI that can use interfaces, documents, and command-line tools may gradually take over repetitive operations. In the short term, the more realistic change is that workflows will be reorganized and broken into smaller tasks, not that entire jobs will disappear immediately.
For the consumer market: Animated images and short-video generation will continue to spread, making marketing assets, production, and personal content creation cheaper. At the same time, issues involving quality, homogenization, and copyright disputes will become more prominent.