Hunyuan (混元) is the AI model family made by Tencent, the Chinese giant behind WeChat, QQ and much of China's gaming industry. It's notable for three things: it's open-weight (released under the Apache 2.0 license), it's broadly multimodal (text, image and 3D generation), and it's distributed to over a billion people through WeChat and Tencent's Yuanbao assistant app. Its flagship, Hunyuan Hy3, is a large open-source mixture-of-experts model.

Hunyuan is made by Tencent, one of the world's largest technology companies and the owner of WeChat (Weixin), QQ and a huge games business. Hunyuan powers Tencent's consumer AI assistant, Yuanbao (元宝), and is served to developers through Tencent Cloud.

Hunyuan Hy3 (Hunyuan 3) is Tencent's flagship model, released in July 2026 after an open-sourced preview in April. It's a mixture-of-experts model with roughly 295 billion total parameters and 21 billion active per query, a context window up to 256K tokens, and multiple 'thinking' depth modes. It's strong on coding, long-form text, reasoning and agent tasks, and it's open-weight under Apache 2.0.

Yes, in several senses. It's free to chat with through Tencent's Yuanbao app and inside WeChat. Because the models are open-weight, you can also self-host them at zero per-token cost. And Tencent's hosted API is cheap — the Hy3 flagship is roughly $0.14 per 1M input tokens and $0.56 per 1M output — among the lower prices in the market.

What Is Hunyuan? Tencent's Open-Source AI Explained (2026)

Q: Is Hunyuan open source?

Yes — and that's one of its defining features. Tencent open-sourced the Hunyuan Hy3 flagship under the Apache 2.0 license, with weights on Hugging Face and ModelScope, and has also released smaller models that can run on a phone, plus open multimodal models for image and 3D generation. Like Alibaba's Qwen, Hunyuan is a model you can download and self-host, not just an API.

Q: Is Hunyuan safe to use?

For everyday use it's fine, with the standard Chinese-model caveats: the hosted app and API run on Tencent's infrastructure in China, subject to Chinese data jurisdiction and content rules on politically sensitive topics. But Hunyuan has a real advantage for sensitive work: because it's open-weight, you can self-host it and keep your data entirely on your own hardware — no Chinese company sees your prompts.

What is Hunyuan — Tencent's open-source AI explained

Hunyuan (混元) is Tencent's AI — and it's one of the more interesting models in China because it combines three things few others do at once: it's open-weight, it's broadly multimodal, and it reaches over a billion people through WeChat. Where DeepSeek is about price and Doubao is about consumer scale, Hunyuan is Tencent's answer built on its own unmatched distribution — the WeChat/Yuanbao ecosystem — while also being a genuinely open model you can download and run yourself. Here's what Hunyuan is and why it matters.

Key takeaways

What it is: Tencent's AI model family, powering its Yuanbao (元宝) assistant and built into WeChat.
Flagship: Hunyuan Hy3 (July 2026) — a 295B-parameter open-source mixture-of-experts model (21B active), 256K context.
Open-weight: released under Apache 2.0 on Hugging Face/ModelScope — self-hostable, like Qwen.
Multimodal: not just text — Tencent open-sources Hunyuan models for image and 3D generation too.
Distribution: reaches WeChat's ~1.4 billion users via the "AI friend" and the Yuanbao app.

Who Makes Hunyuan?

Hunyuan is made by Tencent — one of the largest technology companies on earth, and the owner of WeChat (Weixin), QQ, and a huge chunk of the global games industry. That ownership is the key to Hunyuan's strategy: Tencent doesn't need to buy users, it already has more than a billion of them. Hunyuan powers Tencent's consumer AI assistant, Yuanbao (元宝), and is served to developers through Tencent Cloud.

The WeChat Advantage

The single most important fact about Hunyuan's reach is WeChat. Tencent has integrated Hunyuan across its ecosystem — WeChat, QQ Browser, Yuanbao and more — and in 2025 it added a Yuanbao "AI friend" you can add as a contact inside WeChat itself, putting an AI assistant one tap away for WeChat's roughly 1.4 billion monthly users. It's the same lesson as ByteDance's Doubao: in the consumer market, distribution is everything, and Tencent's is arguably the best in China. The difference is that Hunyuan pairs that reach with something Doubao doesn't — genuine openness.

Most Chinese labs pick a lane: cheap (DeepSeek), open (Qwen), or huge distribution (Doubao). Hunyuan is unusual in trying to be all three at once — open weights, multimodal, and delivered through the largest app in China.

Hunyuan Hy3: The Flagship

Hunyuan's current flagship is Hy3 (Hunyuan 3), released in July 2026 after an open-sourced preview in April. The specs are serious:

Architecture: a mixture-of-experts model with roughly 295 billion total parameters and 21 billion active per query — big capacity, efficient to run.
Context: up to 256K tokens.
Thinking modes: selectable reasoning depth (no-think, low, and high), so you can trade speed for depth.
Strengths: coding, long-form text, reasoning and agent tasks.

Crucially, Hy3 is open-weight under Apache 2.0, with weights on Hugging Face and ModelScope — so you can self-host it, the same way you would an open model like GLM.

More Than Text: Image and 3D

Hunyuan is one of China's most multimodal model families. Beyond the language models, Tencent open-sources Hunyuan models for image generation (its Hunyuan Image line has ranked highly on independent leaderboards) and, notably, 3D generation — turning text or images into 3D assets, an area where Tencent's open models are among the strongest available. It has also released small language models designed to run on a phone. If your work touches image or 3D content and you want open models, Hunyuan is a standout.

Is Hunyuan Free?

Yes, in several ways. You can chat with it for free through the Yuanbao app and inside WeChat. Because the models are open-weight, you can self-host them at zero per-token cost. And Tencent's hosted API is cheap — the Hy3 flagship runs roughly $0.14 per 1M input tokens and $0.56 per 1M output, among the lower prices in the market and well within the range that has defined China's API price war. (Compare it against other models in our Token Calculator.)

How Hunyuan Fits Among the Others

vs Qwen — both are open-weight, but Hunyuan adds Tencent's WeChat distribution and stronger multimodal (3D) models; Qwen is the broader, more-adopted open-weight standard.
vs Doubao — both ride giant consumer platforms (WeChat vs Douyin), but Hunyuan is open and multimodal where Doubao is closed.
vs Ernie — Ernie is the closed, search-integrated incumbent; Hunyuan is the open, WeChat-integrated challenger.

Is Hunyuan Safe?

The hosted app and API run on Tencent's infrastructure in China, so the usual caveats apply: Chinese data jurisdiction, and content rules on politically sensitive topics. But Hunyuan has a real edge for sensitive work — because it's open-weight, you can self-host it and keep your data entirely on your own hardware, with no Chinese company seeing your prompts. That makes it, like Qwen, one of the safer Chinese options if you self-host. The full framework is in are Chinese AI models safe?

Should You Use Hunyuan?

Use Hunyuan if you want a capable open-weight model to self-host, you need strong open image or 3D generation, or you're building inside the Tencent/WeChat ecosystem. Look elsewhere if you want the most widely-adopted open-weight standard (Qwen), the lowest raw text price (DeepSeek), or a Western assistant with strong privacy (Claude). To see the whole Chinese field, start with what is Qwen? and what is Ernie?, and for a pick tailored to your use case, our free picker takes about 30 seconds.

Frequently Asked Questions

What is Hunyuan?

Hunyuan (混元) is Tencent's AI model family — open-weight (Apache 2.0), multimodal (text, image, 3D), and delivered to over a billion people through WeChat and the Yuanbao app. Its flagship is the open-source Hunyuan Hy3 mixture-of-experts model.

Who makes Hunyuan?

Tencent, the owner of WeChat, QQ and a major games business. Hunyuan powers Tencent's Yuanbao assistant and is served to developers via Tencent Cloud.

Is Hunyuan open source?

Yes — a defining feature. Hunyuan Hy3 is open-weight under Apache 2.0 (on Hugging Face/ModelScope), and Tencent also open-sources small on-device models and multimodal models for image and 3D generation.

What is Hunyuan Hy3?

Tencent's flagship, released July 2026 — a mixture-of-experts model with ~295B total / 21B active parameters, up to 256K context, and selectable thinking depths. Strong on coding, long text, reasoning and agents, open-weight under Apache 2.0.

Is Hunyuan free?

Yes — free to chat with via Yuanbao and WeChat, free to self-host (open weights), and cheap on the hosted API (Hy3 ≈ $0.14/$0.56 per 1M input/output tokens).

Is Hunyuan safe to use?

Fine for everyday use, with Chinese-model caveats on the hosted service (China jurisdiction, content rules). But since it's open-weight, you can self-host it and keep data on your own hardware — one of the safer Chinese options if you self-host.