Tencent Hy4 Launched: Vision, Computer Use, Price Explained

Tencent has launched Hy4, the full version of its latest large language model, and says it ranks among the world's leading open-source AI models. The big additions over the earlier Hy4 Preview are vision, computer use and game-engine skills. Pricing has not moved: about $0.90 per million input tokens and $2.70 per million output tokens. Model weights have not been published yet, though Tencent says they are coming "in the coming days".

Quick summary

• Announced November 4, 2026 by Tencent.
• New: image understanding, chart analysis and computer use (command line and graphical interfaces).
• Price: about $0.90 input, $2.70 output and $0.045 cached input per million tokens, same as the Preview.
• Launch offers: WorkBuddy and CodeBuddy are free from November 4 to 15, and Tencent Cloud TokenHub is 50% off through December 31, 2026.
• Weights and system cards: promised "in the coming days", not out yet.

What is new in Tencent Hy4?

Tencent describes three headline upgrades. First, Hy4 can now understand images for the first time, including fine-grained image processing, complex charts and mathematical visual reasoning. Second, it can use a computer: it works with command-line tools and graphical interfaces to carry out everyday tasks, and it received specialist training on software such as Blender and CAD. Third, it has stronger instruction following, long-context comprehension and long-horizon agent behaviour. Tencent also reports that hallucinations fell from 12.95% in the Preview to 7.2%, and errors from unsupported assumptions dropped from 23.1% to 5.6%. These are Tencent's own internal evaluations, so they are claims to treat with care until outside testers repeat them.

Is Hy4 good for game development?

This is the part that matters to gamers and modders. Working with developers from Tencent's Interactive Entertainment Group, the Hy team says the model now handles game logic and multiplayer code across Unreal Engine, Unity and Godot. Against the Preview, Tencent reports gains of 19.1% on its GameDev benchmark and 9.8% on GameLogic. Tencent also says it is integrating Hy4 into products including Honor of Kings and Game of Peace, but did not describe what players will see. Our earlier explainers on Mistral Large 4 and Gemini 4 Carbon show how fast rival models are moving in the same space.

How does Hy4 score against other open models?

The only head-to-head figure in the announcement is a blind evaluation of financial tasks. Tencent's team ran it with Tencent Financial Technology: 213 financial professionals, 153 of them from external institutions, judged 579 real-world tasks. Hy4 scored highest overall.

Two caveats: Tencent ran the test itself, and it covers finance tasks only. On the WorkBuddy product benchmark Tencent reports an average score of 83.18 and a 96.15% pass rate, first place in nine of twelve scenarios in the mainland Chinese market. None of this tells you how Hy4 compares with Claude Opus 5.5 or Gemini 4 Carbon on general work, so wait for independent leaderboards.

How much does Hy4 cost and where can you use it?

Hy4 runs on Tencent Cloud TokenHub, WorkBuddy, CodeBuddy, Yuanbao, WeCom's AI assistant (Dayuan) and ima. Third-party platforms including OpenRouter, Hermes and OpenCode are offering the API, which is the easiest route for developers in India. Standard pricing is about $0.90 per million input tokens, $2.70 per million output tokens and $0.045 per million cached input tokens. Until December 31, 2026, TokenHub charges half of that. A cheaper Hy4 Flash is due soon, and Tencent says it will also bring Hy4 to QQ, WeRead and Tencent News.

Is Hy4 safe to run locally?

Tencent says it trained Hy4 on defensive security data and left offensive exploit material out of training. Its Xuanwu Lab and Keen Security Lab found no significant concerns in the offensive-cyber areas they tested. On biology, Tencent says the model has strong foundational knowledge but is more limited than frontier models at open-ended experimental design. Because the weights are not out, no one outside Tencent can check these safety claims yet.

Should you try it?

If you build apps or write code and want a cheap model with vision and computer use, the free WorkBuddy and CodeBuddy window to November 15 is a no-cost way to test it. If you only need a chat assistant, the benefits are less clear until independent benchmarks arrive. For more launches, see our launch calendar.

Related articles

Anthropic FlashProve Explained: Proving Which AI AnsweredSarvam Prism Explained: India's 44 ms Open Decision ModelMistral Large 4 Open Weights: Benchmarks and What It NeedsDeepSeek Super-agent: Open-Source Rival to Claude Code