Tencent Hunyuan 腾讯混元
Tencent's frontier LLM family powering WeChat & QQ AI
Overview
Hunyuan is Tencent's family of frontier LLMs, deployed across the company's own ecosystem (WeChat features, QQ AI, Tencent Docs, Tencent Meeting) and offered to enterprises via Tencent Cloud. Its current flagship is the open-weight Hunyuan Hy3 (July 2026, a 295B-parameter hybrid fast/slow-thinking MoE with 256K context, Apache 2.0), succeeding Hunyuan 2.0 (December 2025), alongside the Hunyuan-Image and Hunyuan-Video multimodal models.
---
混元是腾讯自研的旗舰级大模型矩阵，已落地于微信、QQ、腾讯文档、腾讯会议等内部产品，并通过腾讯云对外开放。当前旗舰为开源的 Hunyuan Hy3（2026 年 7 月，2950 亿参数、快慢思考混合 MoE，256K 上下文，Apache 2.0），接替 Hunyuan 2.0（2025 年 12 月），并配合 Hunyuan-Image、Hunyuan-Video 多模态模型。
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 1 Sep 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
Use cases
What you can produce with Tencent Hunyuan 腾讯混元
- Generate 5-second 720p video clips from text prompts by running the open-source HunyuanVideo model on your own GPUs.
- Fine-tune HunyuanVideo with LoRAs to lock in a specific visual style or character for production video work.
- Create textured, rigged 3D assets from a text prompt or reference image using Hunyuan3D's hosted studio.
- Call Hunyuan LLMs, including the Hy3 flagship, via Tencent Cloud or OpenRouter to power chat and agent applications.
- Deploy the lightweight HunyuanVideo 1.5 model for video generation on smaller GPU footprints.
- Build agent workflows that use Hy3's tool-calling to complete multi-step office tasks.
- Use Hunyuan-powered features across Tencent's own apps — Yuanbao chat, Tencent Docs and Tencent Meeting — without any setup.
ASEAN Perspective
Tencent Hunyuan 腾讯混元 in Southeast Asia
依托腾讯生态（微信、QQ、会议、文档）的最大私域分发优势，是中国 B 端市场的核心大模型之一。
Hunyuan is Tencent's in-house family of large language and reasoning models (including the T1 reasoning line) that powers its consumer apps and is offered via Tencent Cloud. It is a credible frontier-class model for Chinese-language tasks and multimodal work, and Tencent's open-sourcing of some Hunyuan models adds flexibility for developers who want to self-host.
For an ASEAN audience the practical hurdles are real: access via Tencent Cloud is China-centric, documentation and onboarding are Chinese-first, and accounts plus data handling sit under Chinese jurisdiction, which matters for compliance. English-language developer experience lags Western providers. Strong for Chinese-language and mainland-integrated use cases; SEA teams without a China footprint will generally find OpenAI, Anthropic, or Gemini easier to adopt.
What people say
Tencent entered 2025 seen as second-tier behind Alibaba's Qwen, Moonshot's Kimi and Baidu's Ernie; a string of strong releases reversed that. The July 2026 Hunyuan Hy3 flagship — 295B parameters, 21B active — drew genuinely positive developer reception, and an earlier Hy3 preview topped OpenRouter by token usage in April 2026, a signal of real demand rather than benchmark theatre.
HunyuanVideo was open-sourced in December 2024 at 13 billion parameters, billed then as the largest openly-released video model and trained to beat Runway Gen-3 and Luma 1.6 on text alignment, motion quality and visual fidelity; reviewers in 2026 still rate it the strongest open-weight video generator, with a mature LoRA ecosystem behind it. HunyuanVideo-1.5, in November 2025, made the bigger claim: without an obvious quality trade-off it cut the hardware floor from roughly 45GB of VRAM to about 14GB, low enough for 720p, 121-frame generation on a single RTX 4090. HunyuanImage-3.0, from September 2025, is the image half: an 80B mixture-of-experts model with 64 experts that briefly took #1 on LMArena's text-to-image board, singled out for Chinese-character text rendering — historically every generator's weak point.
The real constraint is not quality, it is the licence. The Tencent Hunyuan Community License defines its Territory as the world excluding the EU, UK and South Korea, so using, modifying or distributing the models in those jurisdictions is not permitted under the standard terms. It imposes a 100-million-MAU commercial gate, forbids using outputs to train a competing model, and layers on a 20-category acceptable-use policy. None of that is unusual for a corporate "open" release, but it is more restrictive than Apache or MIT — check it against your jurisdiction before committing engineering time.
The other criticisms: documentation skews Chinese, the largest models demand serious GPU resources, and Tencent's own assistant Yuanbao — past 50 million daily users — still trails ByteDance's Doubao at home. Hunyuan fits developers and studios who want capable open-weight models, especially video and 3D, that they can self-host and fine-tune. Teams wanting a polished, globally supported hosted API with English-first docs are still better served by US frontier providers.
Summary of public user & expert reviews, compiled by RECATOOLS.
Frequently asked questions
About this listing
This entry was compiled from publicly available data including Tencent Hunyuan 腾讯混元's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Tencent Hunyuan 腾讯混元 unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Tencent Hunyuan 腾讯混元 directly →
Spotted something out of date? Suggest an update →
Alternatives to Tencent Hunyuan 腾讯混元
More in LLMs & Chat
