Tencent Releases Full Hy4 Model With Vision and Computer Use

Tencent announced the full release of Hy4 on November 4, 2026, adding vision and computer-use capabilities and making the model available through hosted APIs and Tencent products. The model weights and a comprehensive system card are planned for a later release.

The earlier Hy4 preview, by contrast, was open-sourced when it arrived in August 2026. It had 770 billion total parameters, 49 billion active parameters and a 1M-token context window, and Tencent identified excessive reasoning and over-verification as known issues in that early version.

The full release substantially scales both pre-training and post-training, with Tencent reporting gains over the preview in instruction following, long-context understanding, long-horizon tasks and deep research.

Hosted Access at Unchanged List Prices

Hy4 is available through Tencent Cloud and OpenRouter and inside CodeBuddy, WorkBuddy, Yuanbao and ima. Listed API pricing is unchanged from the preview at $0.834 per million input tokens, $2.501 per million output tokens and $0.042 per million cached input tokens.

OpenRouter’s deployment listing specifies a 1,048,576-token context window and a 65,536-token completion limit. It accepts text and image inputs and returns text, with Tencent Cloud as its only listed provider. Those specifications make vision part of the hosted API rather than limiting it to Tencent’s own applications.

Vision and Professional Workflows

Tencent says Hy4 can interpret fine-grained image details and complex charts and can operate across command-line and graphical interfaces. Targeted training covered Blender, CAD and long-horizon workflows spanning multiple applications, connecting visual perception with planning and tool use for professional tasks.

Frontend optimization covered complex user-interface interactions, business logic, state management and full-stack integration. Tencent Interactive Entertainment Group developers helped curate game-development training data and evaluations covering Unreal Engine, Unity and Godot, gameplay logic, asset pipelines and multiplayer development. Tencent reports gains of 19.1% on GameDev and 9.8% on GameLogic over the preview.

For finance, Tencent built thousands of realistic workspaces and tens of thousands of tasks, then evaluated Hy4 with 213 finance professionals, including 153 experts from financial institutions, across 579 real-world tasks. In that company-run evaluation, Tencent says Hy4 ranked first among open-weight models with a score of 3.04, ahead of DeepSeek-V4.1-Flash-1017 at 2.97 and Kimi K3.1 Pro and GLM-5.3 at 2.68 each.

Tencent also reports that Hy4 cut hallucinations in internal search evaluations from 15.76% in the preview to 5.45%, and reduced speculative errors from 23.1% to 5.6%. On TenureTrail, an internal set of 123 research tasks across seven disciplines, Hy4 ranked first among open-weight models.

Safety Checks and the Planned Open-Weight Release

Tencent acknowledges that a gap remains to top proprietary models, particularly on challenging internal benchmarks for research-level mathematics, long-horizon coding, model self-improvement and advanced visual question answering.

During training, Tencent limited cybersecurity data to defensive proof-of-concept generation and vulnerability patching, while filtering exploit development, reverse engineering and penetration-testing material. Tencent Xuanwu Lab and Tencent Keen Security Lab conducted red-team audits, and the company reports no elevated risk in reversing cryptographic algorithms, defeating binary obfuscation or exploiting vulnerabilities.

Tencent reports a score of 30.4 out of 100 on Hy-BioExperimentBench, with internal reviewers finding the model’s ability to turn biological knowledge into actionable experimental protocols limited and no evidence that it meaningfully lowers the barrier to biological harm.

The planned weights are to be accompanied by a comprehensive system card covering safety evaluations and additional technical details, with ecosystem support developed alongside the vLLM and SGLang communities. A separate Hy4 Flash release is planned for the near future, targeting significantly improved cost efficiency and faster inference.

The practical result is a broader hosted offering now, while open deployment, fuller safety documentation and the faster, more cost-efficient variant remain future deliverables rather than parts of this release.