# Tencent brings Hy3 global through WorkBuddy, Miora and TokenHub > Source: > Published: 2026-08-05 07:19:44+00:00 [Tencent Cloud](https://x.com/tencentcloud?ref=runtimewire) opened broader international access to [Hy3](https://hy.tencent.ai/?ref=runtimewire) on Wednesday, placing its month-old reasoning model inside the products Tencent expects enterprises and developers to use for office work, creative production and model deployment. The August 5 rollout makes Hy3 available through [WorkBuddy](https://www.workbuddy.ai/?ref=runtimewire), [Tencent Design Miora](https://miora.design/?ref=runtimewire) and [Tencent Cloud TokenHub](https://www.tencentcloud.com/products/tokenhub?ref=runtimewire), alongside API access and downloadable weights. WorkBuddy is free worldwide through August 31, Pacific Time, [Tencent said in its announcement](https://www.prnewswire.com/news-releases/tencent-hy3-now-available-globally-extending-practical-ai-across-products-workflows-and-cloud-services-302843373.html?ref=runtimewire). Poshu Yeung, Tencent Cloud senior vice president and head of Tencent Cloud International, is leading the commercial push. His task is larger than introducing another model endpoint. Tencent is trying to carry Hy3 from early developer traffic into recurring use inside the workflows where companies already create documents, analyze data, write software and produce design assets. That strategy follows the playbook set by Tencent co-founder and CEO [Pony Ma](https://www.tencent.com/zh-hk/who-we-are/our-team/?ref=runtimewire). In Tencent's [first-quarter report](https://static.www.tencent.com/uploads/2026/05/13/59cf8219adbb046153f69387479350ac.pdf?ref=runtimewire), Ma described the cash flow and use cases from Tencent's established businesses as fuel for its AI investments. Hy3 turns that advantage into a product loop: Tencent's applications supply training and product feedback, while improvements to the model flow back into those applications. ### A model built inside the products selling it Hy3 officially launched on July 6, following a preview released in late April. Tencent's [model materials on Hugging Face](https://huggingface.co/tencent/Hy3?ref=runtimewire) describe a Mixture-of-Experts architecture with 295 billion total parameters, 21 billion active parameters and a context window of up to 256,000 tokens. The model targets reasoning, coding, instruction following, data work and multi-step agent tasks. Tencent says Hy3 was refined through feedback from millions of users and large-scale real-world scenarios. The model was co-designed within Tencent's product ecosystem, with the company positioning that usage feedback as a way to improve stability and performance across software development, office productivity, financial modeling, front-end design and game production. Tencent says Hy3 can match open models with two to five times its parameter count. The published comparisons remain Tencent-run evaluations, and the announcement does not provide an independent benchmark table. The architecture also gives Tencent a cost argument. Sparse activation means only a fraction of Hy3's total parameters are used for each token. [OpenRouter's Hy3 listing](https://openrouter.ai/tencent/hy3?ref=runtimewire) starts at $0.1288 per million input tokens and $0.5336 per million output tokens as of August 5, although provider pricing can change. Tencent released the weights under the Apache 2.0 license, allowing developers to adapt and deploy Hy3 in their own environments. The Hugging Face materials include deployment instructions for Transformers, vLLM, SGLang and Docker, giving developers an alternative to Tencent's hosted API. ### Tencent is selling the workflow around the model WorkBuddy is the clearest test of Tencent's product-first approach. The agent workspace can run parallel tasks, use more than 100 built-in skills and accept natural-language instructions for research, data analysis, documents, presentations and workplace communications. [Tencent says in its announcement](https://www.prnewswire.com/news-releases/tencent-hy3-now-available-globally-extending-practical-ai-across-products-workflows-and-cloud-services-302843373.html?ref=runtimewire) that Hy3 achieved a task success rate above 90% in an internal WorkBuddy evaluation and reduced average completion time by 34% against the previous Hy model iteration. Tencent did not publish the task mix, sample size or full methodology, so the figures establish an internal engineering result rather than an independent measure of workplace performance. Miora gives Hy3 a different surface. The creative studio coordinates specialist agents to generate graphics, video, 3D elements and user interfaces from a natural-language brief while preserving an editable production process. Wednesday's announcement puts the finished Hy3 release behind the product; it does not mark Miora's initial launch. TokenHub serves the infrastructure layer. Tencent's model service exposes Hy3 and third-party models through one API, then routes requests according to task, price and performance requirements. That lets Tencent participate even when a customer selects another lab's model. Hy3 can attract usage, while TokenHub remains the control plane. Tencent is extending that distribution through Cafe24 in South Korea and Metelix in Japan, which will offer Hy3 through their respective AI gateway and enterprise platform services. Tencent also named Hermes, Kilo, Cline, OpenClaw, OpenCode and Cherry Studio among the developer platforms adding access. The agent push reaches beyond Tencent's internal product portfolio. [RuntimeWire reported in July](/article/tencent-manus-largest-shareholder-meta-unwind) that Tencent was discussing a deal that could make it the largest shareholder in Manus, Xiao Hong's agent company, after a proposed Meta transaction was blocked. Together, the moves show Tencent pursuing both sides of agent distribution: building models and tools internally while positioning itself around independent agent products. ### Early traffic now faces a conversion test Tencent says Hy3 generated over 68 times the API-call volume of its predecessor and reached the top of OpenRouter's usage leaderboard within a week of its July 6 release, [according to the company's announcement](https://www.prnewswire.com/news-releases/tencent-hy3-now-available-globally-extending-practical-ai-across-products-workflows-and-cloud-services-302843373.html?ref=runtimewire). OpenRouter traffic measures activity on one aggregation platform. It does not establish overall market share or model quality, and a surge following a release can reflect testing, promotional pricing and developer curiosity. The global rollout gives Tencent a clearer commercial test. Free WorkBuddy access can place Hy3 in front of users quickly. Apache-licensed weights can spread the model through developer tools. TokenHub, meanwhile, gives enterprises a managed route with billing, governance and model selection. Sustained adoption will depend on whether those entry points produce workflows that teams keep running after the free period. Tencent has the infrastructure and product reach to make that conversion possible. Hy3 is the model Tencent is using to find out whether its distribution advantage travels beyond China.