<!-- mobian-agent-page publisher="dailydev" canonical="https://daily.dev/posts/tencent-releases-open-weights-for-hy4-a-770b-multimodal-moe-model-tneryyhdt" -->

---
title: Tencent releases open weights for Hy4, a 770B multimodal...
description: Tencent Hunyuan has released open weights for Hy4, its natively multimodal flagship model. It's a mixture-of-experts model with 770B total parameters but only...
canonical: https://daily.dev/posts/tencent-releases-open-weights-for-hy4-a-770b-multimodal-moe-model-tneryyhdt
twitter:card: summary_large_image
twitter:site: @dailydotdev
og:type: website
og:site_name: daily.dev
og:title: Tencent releases open weights for Hy4, a 770B multimodal MoE model | daily.dev
og:description: Tencent Hunyuan has released open weights for Hy4, its natively multimodal flagship model. It's a mixture-of-experts model with 770B total parameters but only...
og:url: https://daily.dev/posts/tencent-releases-open-weights-for-hy4-a-770b-multimodal-moe-model-tneryyhdt
og:image: https://api.daily.dev/og/posts/TNERyyhdT.png
og:image:alt: Tencent releases open weights for Hy4, a 770B multimodal MoE model
og:image:width: 1200
og:image:height: 630
og:locale: en
---

> ## Documentation Index
> Fetch the complete documentation index at: https://daily.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Tencent releases open weights for Hy4, a 770B multimodal MoE model

**[Collections](https://daily.dev/sources/collections)** · 1 min read · 1 upvotes · 0 comments

## Summary

Tencent Hunyuan has released open weights for Hy4, its natively multimodal flagship model. It's a mixture-of-experts model with 770B total parameters but only 49B active per token, about 6% of weights, and a 1M token context window. That sparsity means inference should cost closer to a much smaller dense model, though serving the full weights still needs serious hardware. Only the base model is out so far; benchmarks, licensing terms and training data details weren't part of the announcement.

## Content

Tencent Hunyuan has released open weights for Hy4, its natively multimodal flagship model. The company describes it as a mixture-of-experts (MoE) model with 770B total parameters and 49B active per token, and a context window of 1M tokens.

The Hy4 base model is available on Hugging Face. The release was announced by @TencentHunyuan and then amplified by Hugging Face-focused accounts, including @HuggingPapers, which called it a multimodal foundation model, and by @mervenoyann and @_akhaliq.

Only about 6% of the parameters (49B of 770B) are active for any given token, so inference cost should be closer to that of a much smaller dense model. The full weights are still large, though, and serving them will take serious hardware.

What's published so far is the base model. Benchmarks, licensing terms and details of the training data weren't part of the announcements.

## Questions this post answers

### How many parameters does Tencent's Hy4 model actually use per token?

Hy4 has 770 billion total parameters but only 49 billion are active per token, roughly 6% of the full model. It's a mixture-of-experts architecture with a 1 million token context window, released as open weights on Hugging Face by Tencent Hunyuan. Because so few parameters activate per token, inference cost should resemble a much smaller dense model, though the full weights still require serious hardware to serve.

_daily.dev helps engineers comparing MoE models track open-weight releases like this one as they land._

### Is Tencent's Hy4 model released with benchmarks and a license?

No, only the base model weights have been published so far. Benchmarks, licensing terms, and training data details were not included in the initial announcement from Tencent Hunyuan, so anyone evaluating Hy4 for production use currently lacks the performance data and legal terms needed to make that call.

_Teams deciding whether to adopt a new open-weight model can follow licensing and benchmark updates on daily.dev._

---

Tags: [#open-source](https://daily.dev/tags/open-source), [#mixture-of-experts](https://daily.dev/tags/mixture-of-experts), [#huggingface](https://daily.dev/tags/huggingface)

[View this post on daily.dev](https://daily.dev/posts/tencent-releases-open-weights-for-hy4-a-770b-multimodal-moe-model-tneryyhdt)

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]}
{"@context":"https://schema.org","@type":"TechArticle","headline":"Tencent releases open weights for Hy4, a 770B multimodal MoE model","url":"https://daily.dev/posts/tencent-releases-open-weights-for-hy4-a-770b-multimodal-moe-model-tneryyhdt","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/tencent-releases-open-weights-for-hy4-a-770b-multimodal-moe-model-tneryyhdt"},"datePublished":"2026-11-20T07:52:45.554Z","dateModified":"2026-11-20T07:53:18.253Z","description":"Tencent Hunyuan has released open weights for Hy4, its natively multimodal flagship model. It's a mixture-of-experts model with 770B total parameters but only...","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/tencent-releases-open-weights-for-hy4-a-770b-multimodal-moe-model-tneryyhdt","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"open-source,mixture-of-experts,huggingface","timeRequired":"PT1M"}
{"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"Tencent releases open weights for Hy4, a 770B multimodal MoE model"}]}
{"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/tencent-releases-open-weights-for-hy4-a-770b-multimodal-moe-model-tneryyhdt#faq","mainEntity":[{"@type":"Question","name":"How many parameters does Tencent's Hy4 model actually use per token?","acceptedAnswer":{"@type":"Answer","text":"Hy4 has 770 billion total parameters but only 49 billion are active per token, roughly 6% of the full model. It's a mixture-of-experts architecture with a 1 million token context window, released as open weights on Hugging Face by Tencent Hunyuan. Because so few parameters activate per token, inference cost should resemble a much smaller dense model, though the full weights still require serious hardware to serve. daily.dev helps engineers comparing MoE models track open-weight releases like this one as they land."}},{"@type":"Question","name":"Is Tencent's Hy4 model released with benchmarks and a license?","acceptedAnswer":{"@type":"Answer","text":"No, only the base model weights have been published so far. Benchmarks, licensing terms, and training data details were not included in the initial announcement from Tencent Hunyuan, so anyone evaluating Hy4 for production use currently lacks the performance data and legal terms needed to make that call. Teams deciding whether to adopt a new open-weight model can follow licensing and benchmark updates on daily.dev."}}]}
```

