Tencent launches Hy4 with vision and computer use, but the open weights are still to come

Tencent announced on 4 November 2026 the release of Hy4, its latest large language model, which succeeds the Hy4 Preview.

What is new

• Visual understanding for the first time: image processing, chart analysis and visual maths reasoning.
• Computer use for the first time: the model can operate both command-line and graphical interfaces. Tencent says it was specially trained on professional software such as Blender and CAD.
• Improvements in instruction following, long-context comprehension and long-horizon agent tasks.

Tencent’s own figures

All benchmark numbers come from Tencent and have not been independently verified:

• In a blind evaluation of 579 financial tasks with 213 financial professionals, Hy4 scored 3.04, ahead of DeepSeek-V4.1-Flash (2.97), Kimi K3.1 Pro (2.68) and GLM-5.3 (2.68).
• In internal tests the hallucination rate fell from 12.95% to 7.2% compared with the preview.

Availability and price

Hy4 is available in Tencent products including WorkBuddy, CodeBuddy, Yuanbao and Tencent Cloud TokenHub, and via third-party platforms such as OpenRouter. Pricing is unchanged from the preview: about $0.90 per million input tokens and $2.70 per million output tokens. TokenHub offers 50% off through 31 December 2026.

Open source — not yet

Tencent describes Hy4 as among the world’s leading open-source models, but says the weights and system cards will be released “in the coming days”. Until then the model can only be used via API, the NPDev briefing notes. Tencent says its safety work focused on cybersecurity and biosafety, and that internal assessments found no significant concerns about offensive cyber capabilities in the areas evaluated.