
Hy4
Tencent · China · #4 in our ranking · Best for computer use
Hy4 is an open-weight mixture-of-experts model from Tencent (China), released Nov 2026, with 770B parameters (49B active per token), a 1M-token context window, and the Tencent Hy License 1.0.
Hy4 is a mixture-of-experts model with 8 of 256 routed experts plus a shared expert, sparse attention, and a separate vision encoder. It reads text, images, and video and supports a 1M-token context. The Tencent Hy License is based on Apache 2.0 but adds commercial terms for large hosted providers, so read it before you deploy.
Hy4 specs
- Provider
- Tencent
- Origin
- China
- Architecture
- Mixture-of-experts
- Parameters
- 770B total · 49B active
- Context
- 1M tokens | 786K words
- License
- Tencent Hy License 1.0
- Input
- Text, Image, Video
- Released
- Nov 2026
- Weights
- Hugging Face
Sourced from the model card on Hugging Face. Last checked 2026-11-21.
Highlights
- Reads text, images, and video
- OSWorld 2.1 score of 78.9 in Tencent's report
- Custom license with terms for large hosted providers
Where it fits for payment teams
Worth testing for agents that work through back-office web portals, where computer use matters more than raw coding.
Harnesses in Runtime
Run Hy4 through any of these harnesses on Runtime.
Provider-reported benchmarks
| Terminal-Bench 4.0 | 44.4 |
|---|---|
| DeepSWE v1.1 | 74.9 |
| OSWorld 2.1 | 78.9 |
Benchmark scores are reported by each model provider on its Hugging Face model card. They are not independently verified by Runtime.
How to run Hy4
GPU requirements, Bedrock, and Vertex AIDownload the weights
Pull tencent/Hy4 from Hugging Face. Check the Tencent Hy License 1.0.
Serve it in your cloud
vLLM on your GPUs, or a managed host.
Put it to work in Runtime
Point an agent or skill at it, then compare with evals.
Hy4 FAQ
What is the context window of Hy4?
+
Hy4 supports a 1M-token context window (1,048,576 tokens), roughly 786K words of English text.
How many parameters does Hy4 have?
+
Hy4 is a mixture-of-experts model with 770B total parameters, of which 49B are active per token.
Is Hy4 free for commercial use?
+
Hy4 is released under the Tencent Hy License 1.0, a custom license. Commercial use may come with conditions, so read the license on Hugging Face before you deploy or fine-tune it.
What can Hy4 take as input?
+
Hy4 accepts text, image, video input and generates text.
Can I run Hy4 in my own cloud with Runtime?
+
Yes. Download the weights from Hugging Face (tencent/Hy4) and serve them with vLLM in your cloud, or use a managed provider that hosts the model. Runtime agents can then use Hy4 through a harness like OpenCode, with your data staying in your environment. Our how-to-run page for Hy4 covers GPU memory, Bedrock, and Vertex AI.