Tencent is taking its latest artificial intelligence model outside China in a much bigger way.
The technology giant has expanded global access to Tencent Hy3 AI, its newest large language model, opening it to developers, businesses and everyday users through several of the company’s cloud and productivity products.
Hy3 was officially released in July and is now available internationally through WorkBuddy, Tencent Design Miora and Tencent Cloud TokenHub. Tencent says more integrations across its AI-focused cloud products are coming.
This isn’t simply Tencent putting another chatbot on the internet. The more interesting part is where Hy3 is showing up — inside coding environments, workplace agents, creative software and enterprise infrastructure.
Tencent Hy3 Is Built for More Than Chat
Under the hood, Hy3 uses a hybrid Mixture-of-Experts (MoE) architecture.
The model contains 295 billion total parameters, but only 21 billion are activated during inference. It also supports context windows reaching 256,000 tokens. Tencent is positioning that architecture as a way to deliver strong reasoning and coding performance without requiring the computing resources associated with activating an entire model of that size for every request.
The company is particularly interested in agentic AI. Rather than waiting for a user to ask one question after another, AI agents can be given a broader objective and work through multiple steps themselves. Tencent says Hy3 has been designed around reasoning, coding and these more autonomous workflows.
That distinction matters. The AI industry is steadily moving away from models that merely generate answers toward systems expected to actually complete work.
WorkBuddy Gives Hy3 a Workplace Test
One of the clearest examples is WorkBuddy, Tencent’s AI workplace platform. Hy3-powered agents inside WorkBuddy can take on research, document creation, presentations, coding and workplace communications from a single prompt.
Tencent says its internal testing found a 34% improvement in task completion time, while task success rates exceeded 90%. Those are company-reported figures rather than independent benchmark results, but they show where Tencent sees the model’s value: completing workflows, not just winning chatbot comparisons.
Global users can access Hy3 through WorkBuddy free until August 31, 2026, Pacific Time. That gives Tencent a relatively straightforward way to put the model in front of international users without asking them to immediately set up APIs or cloud infrastructure.
Developers Aren’t Locked Into Tencent’s Own Apps
Tencent is also pushing Hy3 directly toward developers. The model can be connected through APIs and used with developer environments, coding extensions and third-party tools. Integrations already include platforms such as Hermes, Kilo, Cline, OpenClaw, OpenCode and Cherry Studio.
Hy3 is also available through open model channels under the Apache 2.0 license, giving developers considerably more flexibility than a closed, API-only release. Tencent says the model has been available through open-source model communities since launch.
That could matter more than another benchmark victory. Developers tend to follow models they can actually experiment with, modify and plug into existing workflows. Open distribution also gives Tencent another route into the international AI developer ecosystem, where OpenAI, Anthropic, Google, Meta, Alibaba, DeepSeek and a growing list of open-model developers are fighting for attention.
Hy3 Is Already Getting Heavy API Usage
Early usage numbers suggest Hy3 hasn’t had a quiet launch. Tencent says the model generated 68 times more API calls than its predecessor and reached the top of OpenRouter’s global large language model usage rankings within a week of release.
The company has also put aggressive pricing behind the rollout. According to Tencent figures reported by Daily Tribune, Hy3 pricing on OpenRouter starts at $0.1288 per million input tokens and $0.5336 per million output tokens.
For developers running large numbers of requests, that kind of pricing can become part of the decision just as quickly as benchmark scores. A model doesn’t necessarily need to be the absolute best at every task if it’s capable enough, cheap enough and easy to deploy.
Tencent Is Connecting Hy3 With Its Creative AI Tools
Hy3 isn’t staying inside coding and office software either. Tencent has integrated the model into Tencent Design Miora, its AI-powered creative environment for producing graphics, video, interfaces and 3D content.
There’s a broader strategy visible here. Tencent has already been investing heavily in generative 3D technology. Its Hunyuan 3D ecosystem supports AI-generated 3D assets from inputs including text, images and sketches, while newer systems extend that work toward more sophisticated 3D generation and editing.
Hy3 gives Tencent a general-purpose intelligence layer that can sit alongside those specialized models rather than replacing them.
Tencent Cloud Wants Hy3 Inside Enterprise AI Stacks
For companies juggling several AI providers, Tencent is offering another route through Tencent Cloud TokenHub.
TokenHub is designed to let enterprises access and manage different AI models through a unified API instead of building separate integrations for every provider. Hy3 now becomes one of the models available within that infrastructure. It’s a less flashy part of the announcement, but potentially one of the more important ones.
Enterprise AI adoption increasingly involves multiple models. A business might use one model for coding, another for document processing and another for complex reasoning. Platforms capable of routing and managing that mixture have become their own competitive layer. Tencent clearly wants a position there.
China’s AI Competition Is Becoming Global
Hy3’s international expansion arrives while Chinese technology companies are becoming much harder to ignore in the global AI model race. Tencent is no longer treating Hunyuan — now branded Tencent Hy — as something primarily meant for its domestic ecosystem. The company has moved quickly from rebuilding its model infrastructure in early 2026 to releasing the Hy3 preview in April and the full model in July.
Now comes the harder part: convincing developers outside China to keep using it. Low pricing helps. Open-source availability helps more. Putting the model inside products people can immediately try may help most of all.
Hy3 doesn’t need to replace every established AI model to make Tencent relevant internationally. If developers start keeping it in their model stack because it’s inexpensive, open and capable enough for real workloads, Tencent has already gained something valuable.
Global distribution turns Hy3 from another major Chinese AI release into a model that has to compete for actual usage. And that’s a much more interesting test.
