What it does
Hy4 is the Tencent Hy (Hunyuan) team's flagship language model, released on 4 November 2026 as the full version of Hy4 preview. It is the first Hy model with vision: it reads fine-grained image detail and complex charts, and it can operate a computer through both the command line and graphical apps, including professional software such as Blender and CAD. Tencent also trained it for frontend and game development, finance and office documents, scientific research and long-running assistant work. It is available through the Tencent Cloud and OpenRouter APIs and inside Tencent's CodeBuddy, WorkBuddy, Yuanbao and ima apps; Tencent says the open weights will follow later.
Official source 1 ↗Features
- Vision and chart reading. Hy4 is the first Hy model that accepts images. It combines perception, reasoning and tool use to pick out fine-grained image details and interpret complex charts. Official source 1 ↗
- Computer use across CLI and GUI. Hy4 can operate a computer through both command-line tools and graphical interfaces. Tencent trained it on professional software such as Blender and CAD and on long workflows that move between several applications. Official source 1 ↗
- Frontend and game development. Tuned for complex UI interactions, business logic, state management and full-stack integration, and adapted for Unreal Engine, Unity and Godot, including multiplayer game development. Official source 1 ↗
- Finance and office documents. Trained on finance tasks such as fund research, equity analysis and corporate finance, with search upgrades that cut the hallucination rate in Tencent's internal tests from 15.76% to 5.45%. It can co-edit Word, Excel and PowerPoint files with people in real time. Official source 1 ↗
- Scientific research. Tencent reports that Hy4 ranks first among open-weight models on TenureTrail, its internal benchmark of 123 research tasks across seven disciplines. Official source 1 ↗
- Long-running personal assistant. In WorkBuddy with WeCom, Hy4 draws on chats, emails, documents, calendars and meeting notes to track projects, surface action items, prepare meeting briefs and flag deadlines. Official source 1 ↗
Pricing & free limits
Cached input is $0.042 per 1M tokens. Same API price as Hy4 preview; available on Tencent Cloud and OpenRouter. Official source 1 ↗
Billing details
Billing details were not established by the checked official sources.
Free access & limits
Free access and its limits were not established by the checked official sources.
Best for & limitations
Running agents that operate desktop software, such as Blender or CAD, or work across several applications in one task. Official source 1 ↗
Building production-ready web frontends and games for Unreal Engine, Unity or Godot. Official source 1 ↗
Finance research and office work: equity analysis, valuation models and editing Word, Excel and PowerPoint files. Official source 1 ↗
Reading screenshots, detailed images and complex charts as part of a reasoning or tool-use task. Official source 1 ↗
- Open weights not yet released. Tencent calls Hy4 an open-source model, but the weights are not out yet: it plans to release them, with a system card, later. Today Hy4 is available only through the API and Tencent's apps. Official source 1 ↗
- Behind top proprietary models. Tencent says a notable gap remains to the top proprietary models on research-level math, long-horizon coding, model self-improvement and advanced visual question answering. Official source 1 ↗
- Benchmarks are Tencent's own. Most results Tencent cites, such as TenureTrail and the hallucination figures, come from internal evaluations rather than independent tests. Official source 1 ↗
- Restricted cybersecurity training. Tencent limited Hy4's cybersecurity training data to defensive material such as proof-of-concept generation and vulnerability patching. Official source 1 ↗