Tencent released the full version of Hy4 on November 4, 2026, adding vision and computer-use capabilities to its flagship model and making it available through hosted APIs and Tencent’s own products at unchanged prices. Why it matters: Hy4 is one of China’s largest frontier models, and this release moves it from a text-only preview into the multimodal, agent-style work that Western labs are also racing to own.
November 4, 2026
What happened
Hy4 first arrived as a preview in August 2026, when Tencent open-sourced a model with 770 billion total parameters, 49 billion active parameters and a 1M-token context window — while flagging excessive reasoning and over-verification as known issues in that early version. The full release announced on November 4 substantially scales both pre-training and post-training, and Tencent reports gains over the preview in instruction following, long-context understanding, long-horizon tasks and deep research.
The headline additions are new senses and hands. For the first time in the Hy line, the model can interpret fine-grained image details and complex charts, and it can operate across both command-line and graphical interfaces. Tencent says targeted training covered professional workflows including Blender, CAD and multi-application tasks, plus game-development work in Unreal Engine, Unity and Godot.
Availability and pricing
Hy4 is available now through Tencent Cloud and OpenRouter, and inside Tencent’s CodeBuddy, WorkBuddy, Yuanbao and ima products. Listed API pricing is unchanged from the preview: $0.834 per million input tokens, $2.501 per million output tokens and $0.042 per million cached input tokens. OpenRouter’s listing specifies a 1,048,576-token context window and a 65,536-token completion limit, with text and image inputs.
One important caveat: the full release is hosted-only for now. Tencent says the model weights and a comprehensive system card are planned for a later release, and a separate, cheaper and faster Hy4 Flash variant is planned for the near future. Neither is part of today’s launch.
Safety, in Tencent’s telling
Tencent says it deliberately limited the model’s offensive cybersecurity training data — filtering out material on exploit development, reverse engineering and penetration testing while keeping defensive content — a step Reuters noted follows a wave of AI-driven cyberattacks and a similar restriction by Moonshot AI last month. The company also reports that internal reviewers found no evidence Hy4 meaningfully lowers the barrier to biological harm, and that it cut hallucinations in its internal search evaluations from 15.76% with the preview to 5.45%. Those are company-reported figures, and Tencent itself acknowledges a remaining gap to top proprietary models on its hardest internal benchmarks.
Why it matters
For readers tracking the open-model race, Hy4’s full release matters less for any single benchmark than for what it signals: China’s biggest platform companies are shipping frontier-scale models with vision and computer use built in, at API prices well under a dollar per million input tokens, while explicitly designing around cyber-misuse fears. The honest test comes later — when the weights and system card actually land, outside developers can verify the claims. Until then, Hy4 is a hosted product with big stated ambitions, and that is exactly how we are treating it: everything AI, tested — once there is something public to test.
FAQ
Is Tencent Hy4 open source?
Not yet for the full release. The August preview was open-sourced, but Tencent says the full Hy4 weights and a system card are planned for a later release. Today it is available through hosted APIs and Tencent products.
What is new in the full Hy4 release?
Vision (image and chart understanding) and computer use across command-line and graphical interfaces, on top of scaled-up training that Tencent says improves instruction following, long-context work and deep research over the preview.
How much does Hy4 cost?
Listed API pricing is unchanged from the preview: $0.834 per million input tokens, $2.501 per million output tokens and $0.042 per million cached input tokens.
Sources: Reuters, Unite.AI (reporting Tencent’s announcement), OpenRouter listing