GitHub radar
Tencent Hy4: open 770B with 1M context
Tencent has launched a preview version of Hy4, a 770B parameter MoE model with a 1M token context window and an Apache 2.0 license. It supports tasks such as coding, office work, game development and scientific reasoning.
The Hy4 preview version is a next-generation MoE flagship model developed by the Tencent Hy Team. It has 770 billion parameters and activates 49 billion parameters per token at inference time. This allows it to achieve the performance of a large model without having to compute all of its weights at once. It also has a 1M token context window, which is approximately equivalent to 750,000 words. This allows the model to process entire codebases or books at once. It also supports tasks like software engineering (such as understanding, planning and debugging long horizon development tasks), office work (including converting context into various types of documents, spreadsheets and financial models), game development (creating playable prototypes from textual descriptions) and scientific reasoning (in AI, physics and mathematics). It is licensed under the Apache 2.0 license. To run it, you will need 8 tensor parallel GPUs. It is a server side model.
Why a vibe-coder should care
With an Apache 2.0 license, any cloud service provider can host the Hy4 model and provide an API. So we might see different cloud services offering competitive pricing on a per token basis very soon. A 1M token context window is pretty useful as well since you can send your whole codebase or document at once without needing to do any workarounds.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Find cloud providers that already offer API access to Hy4 from Tencent: https://huggingface.co/tencent/Hy4-preview — compare pricing and show me how to make a basic API call.
This is a server-grade model — you can't run it at home, but it sets the bar.
Open on Hugging Face▌ More finds