Tencent Expands Hy3 Globally Across Agents, APIs and Open Platforms

The August update expands distribution rather than launching Hy3 again

Tencent announced a broader global rollout for Hy3 on August 5, 2026. The model itself was first released on July 6, so this update should be understood as an international access and ecosystem expansion rather than a second model launch.

Hy3 is a Mixture-of-Experts model with roughly 295 billion total parameters and 21 billion active parameters per token. Tencent describes it as combining fast and slow thinking modes, supporting a 256K context window, and making its weights available under the Apache 2.0 license.

Access now spans agents, design tools, cloud APIs and open platforms

Tencent is distributing Hy3 through several product layers. WorkBuddy uses the model for agentic workplace tasks. Design Miora applies it to design-oriented workflows. Tencent Cloud TokenHub exposes hosted access for developers, while the weights are also available through open model communities and compatible deployment ecosystems.

The significance is not a new architecture change but broader availability: one model can be used through finished agent products, managed APIs or self-hosted open-weight deployments depending on the user's operational requirements.

Tencent reports strong internal agent results, but they remain vendor measurements

Tencent says internal WorkBuddy evaluations achieved more than 90% task success and completed work 34% faster than the previous Hy model. Those figures come from Tencent's own evaluation environment and should not be treated as universal performance guarantees for unrelated enterprise tasks.

The company also reports that Hy3 API-call volume grew to more than 68 times that of the previous model and that Hy3 became one of the most-used models on OpenRouter shortly after launch. Usage rankings can change quickly and are not equivalent to independent quality benchmarks.

Tencent cites starting OpenRouter prices of approximately $0.1288 per million input tokens and $0.5336 per million output tokens at the time of its announcement. Third-party pricing is dynamic, so developers should verify current rates before estimating production costs.

Global infrastructure is part of the rollout story

Tencent says its cloud footprint spans 66 availability zones across 23 regions and more than 3,200 acceleration nodes, with regional partners including providers in Japan and South Korea. These figures describe Tencent's broader infrastructure footprint rather than dedicated Hy3-only capacity.

For organizations comparing deployment options, the practical decision is therefore broader than model quality. They can weigh hosted regional access, agent-product integration, open-weight self-hosting, latency, data residency and cost.

The most important freshness distinction remains simple: Hy3 was released in July; August 5 expanded how and where users can access it globally.