Tencent Hunyuan open-sources Hy4 preview, a 770B-parameter MoE LLM with 1M context
Tencent Hunyuan released and open-sourced Hy4 preview, a new frontier large language model with 770B total parameters, 49B active parameters, and a 1M token context window, available on Tencent Cloud, GitHub, and Hugging Face under the Apache 2.0 license. The model, a major upgrade from Hy3 preview (295B total, 21B active, 256K context), targets coding, office, gaming, and scientific research, and includes optimizations that improved end-to-end throughput by 31.8% over baseline. API pricing is set at 6 yuan per million input tokens and 18 yuan per million output tokens, with a two-week free trial on WorkBuddy and CodeBuddy.
Coverage timeline
腾讯混元 BlogHy LLM
@@@coverCard type = "image" date = "2026年8月28日" media_src = " https://hy-model-ap-prod-1258344703.cos.ap-guangzhou.myqcloud.com/llm-blog/default/9a5ea02f/2026082716220649_1437ddb3b346a01a5e603f2a607d250c.png?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1787818929;2103178929&q-key-time=1787818929;2103178929&q-header-list=&q-url-param-list=&q-signature=87337c8f948a5b795d1bafd4676b33fbba6995fe " image_alt = "hy4PreviewBg" title = "Introducing Hy4 preview" aspectRatio = "" mediaWidth = "100%" mediaMarginTop = "32" mediaMarginBottom = "64" videoLoop = 0 @@@ @@@space height = 16 @@@ @@@buttonGroupCard [[buttons]] text = "Tencent Cloud" icon = "tencentcloud" url = " https://console.tencentcloud.com/tokenhub/models/detail?modelId=hy4-preview " [[buttons]] text = "GitHub" icon = "GitHub" url = " https://github.com/Tencent-Hunyuan/Hy4-preview " [[buttons]] text = "HuggingFace" icon = "huggingface" url = " https://huggingface.co/tencent/Hy4-preview " [[buttons]] tex
机器之心机器之心
今天， 腾讯混元发布并开源新一代大语言模型 Hy4 preview 。 该模型总参数达到 770B，单次推理激活 49B，支持 1M 上下文。代码、办公、游戏和科学研究，是这次升级重点覆盖的场景。 Hy4 preview 还首次参与混元自身的训练方法、数据策略、评估体系和底层算子优化。在一轮推理基础设施优化中，模型分析系统瓶颈并围绕算子融合、通信优化持续迭代，最终将端到端吞吐较基线提升 31.8%。 这是混元今年 2 月重建基础设施后推出的又一个大版本。与上一代 Hy3 preview 的 295B 总参数、21B 激活参数和 256K 上下文相比，Hy4 preview 在模型容量和上下文窗口上都跨出了一大步。 Hy4 preview 的模型权重及 FP8 版本已经开放，采用 Apache 2.0 许可 。用户也可以在 WorkBuddy、CodeBuddy 国内版及国际版、元宝和 ima 中体验，并通过腾讯云 TokenHub、OpenRouter 调用 API。 官方博客：https://hy.tencent.ai/research/hy4-preview Hugging Face 开放权重：https://huggingface.co/tencent/Hy4-preview GitHub 代码：https://github.com/Tencent-Hunyuan/Hy4-preview 至于用户关心的 API 价格， 输入为每百万 tokens 6 元，输出为每百万 tokens 18 元 ，缓存命中输入为每百万 tokens 0.3 元。另外，WorkBuddy 和 CodeBuddy 还将 开启为期两周的限时免费体验 。 进入当前开源模型的第一梯队 根据腾讯公布的模型卡，Hy4 preview 采用混合专家架构，主干共有 78 层。除第一层使用标准稠密前馈网络外，其余 77 层均为 MoE 层，每层包含 256 个路由专家和 1 个共享专家，每个 token 会调用其中 8 个路由专家及共享专家。 这套设计让模型拥有 770B 的总容量，同时把单次计算涉及的参数控制在 49B。模型还内置一层原生 MTP，用于推测解码。注意力模块采用 Gated DSA，并通过 IndexCache 复用跨层稀疏索引；残差通路引入 iHC，扩展层间信息流。这些设计分别指向
Techmeme
Bloomberg : Tencent releases Hy4 Preview, a 770B-parameter foundational model with a 1M context window that it says outperforms Z.AI and Moonshot models in internal tests — Tencent Holdings Ltd. released a foundation model it says outperforms rivals Z.AI Co. and Moonshot AI in internal tests …
Simon Willison
Introducing Hy4 Preview New open weight text input (no vision) LLM from Chinese company Tencent today: 770B total parameters, 49B active parameters, 1M token context window, 1.56TB on Hugging Face . This is a big size increase from their previous Hy3 in July, which was 295B, 21B active, 256,000 context, 598GB. I recently started using model chat templates to better understand their capabilities. Here's Hy4's chat_template.jinja on Hugging Face, which includes this section: {% - if not reasoning_effort is defined %} {% - set reasoning_effort = 'high' %} {% - elif reasoning_effort not in [ 'high' , 'no_think' ] %} {% - if reasoning_effort is none %} {{- raise_exception('reasoning_effort error : None, should be no_think/high') }} {% - else %} {{- raise_exception('reasoning_effort error : ' + reasoning_effort + ', should be no_think/high') }} {% - endif %} {% - endif %} So it looks like there are just two reasoning effort levels: "high" (the default) and "no_think" (reason by disabled). I
