---
format: "aidr-story-markdown/v1"
id: "d40a493d10f5eafe9c72fa27a7e058410031586bc2048b1ab36f91823309016f"
canonical_url: "https://aidr.today/d40a493d?lang=en"
title: "Tencent Open Sources 770B Parameter Hy4 AI Model With 1 Million Token Context"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-09-01T11:47:30.000Z"
category: "Releases"
topics: ["tencent","llm","open-source","inference","moe"]
source_urls: ["https://x.com/TencentHunyuan/status/2094730117541155001","https://x.com/sudeepsriv/status/2094714495872778331"]
summary: "Tencent's Hunyuan team released the Hy4 preview open-weight artificial intelligence model using a Mixture-of-Experts architecture. The model features 770 billion total parameters, 49 billion active parameters per token, and a context window of 1 million tokens. The system identified its own inference bottlenecks during research and increased end-to-end throughput 31.8% via operator fusion and communication optimizations. Designed for coding and productivity, the model employs 256 routed experts as part of a trend of open releases from Chinese developers including DeepSeek, Qwen, and Kimi. Tencent distributed the weights via HuggingFace and Github to allow developers and researchers to build on the architecture."
---

# Tencent Open Sources 770B Parameter Hy4 AI Model With 1 Million Token Context

> [Open the canonical story](<https://aidr.today/d40a493d?lang=en>)

**Published:** 2026-09-01T11:47:30.000Z
**Category:** Releases
**Topics:** tencent, llm, open\-source, inference, moe

## Summary

Tencent's Hunyuan team released the Hy4 preview open\-weight artificial intelligence model using a Mixture\-of\-Experts architecture\. The model features 770 billion total parameters, 49 billion active parameters per token, and a context window of 1 million tokens\. The system identified its own inference bottlenecks during research and increased end\-to\-end throughput 31\.8% via operator fusion and communication optimizations\. Designed for coding and productivity, the model employs 256 routed experts as part of a trend of open releases from Chinese developers including DeepSeek, Qwen, and Kimi\. Tencent distributed the weights via HuggingFace and Github to allow developers and researchers to build on the architecture\.

## Sources

- [Story source](<https://x.com/TencentHunyuan/status/2094730117541155001>)
- [Supporting source](<https://x.com/sudeepsriv/status/2094714495872778331>)