Hy4 Preview (Tencent Hunyuan 4, 腾讯混元Hy4) is Tencent's next-generation Mixture-of-Experts flagship model — the successor to Hy3, built for complex reasoning, coding, and agentic tasks.
Model ID: tencent/hy4-preview · Available via OpenAI-compatible API
Announced by Tencent in its Q2 2026 earnings call and now in grey-testing, Hy4 is positioned as an "expert-level" model above Hy3 — with larger parameters, stronger reinforcement learning, and upgraded multimodal capabilities. It serves as the core engine of Tencent's WorkBuddy agent application. According to the official research page, Hy4 Preview is a new-generation Mixture-of-Experts (MoE) flagship model developed by the Tencent Hy Team.
The model comprises 770B total parameters, of which 49B are activated per token. Its backbone consists of 78 layers:
In addition to the backbone, 1 native MTP layer (Multi-Token Prediction, 10B total parameters, 0.7B activated) is built in for speculative decoding — accelerating inference by predicting multiple tokens ahead.
| Architecture Detail | Value |
|---|---|
| Total / Active parameters | 770B / 49B per token |
| Backbone layers | 78 |
| Layer 1 | Dense FFN |
| Layers 2–78 | MoE (256 routed + 1 shared expert each) |
| Activated experts per token | Top-8 routed + shared |
| MTP layer | 1 native (10B / 0.7B active), speculative decoding |
| Specification | Value |
|---|---|
| Architecture | Mixture-of-Experts (MoE) |
| Total / Active parameters | 770B / 49B |
| Backbone | 78 layers (1 dense + 77 MoE) |
| Context window | 1,048,576 tokens |
| Maximum output | 64,000 tokens |
| Input price | $0.834 / 1M tokens |
| Output price | $2.501 / 1M tokens |
| Provider | Tencent Hy Team |
| Official research page | hy.tencent.ai/research/hy4-preview |
Hy4 targets complex reasoning, coding, and agent scenarios — competing directly with DeepSeek, Kimi, and GPT-class frontier models. Full benchmark comparison on our Hy4 vs DeepSeek page.
| Model | Type | Positioning |
|---|---|---|
| Hy4 Preview | Tencent Hunyuan expert MoE | Complex reasoning / coding / agents |
| Hy3 Preview | Tencent Hunyuan MoE | General-purpose |
| DeepSeek V-series | MoE | Reasoning / coding |
| Kimi / GPT-class | MoE / dense | General assistant |
Hy4 Preview is currently served as a hosted API — the model ID is
tencent/hy4-preview on OpenRouter. It is OpenAI-compatible, so you can call it
through any gateway that supports the OpenAI API format. Create a free account at
tokenra.io/register
to start calling the model right away. See API docs for examples.
Hy4 targets complex reasoning, coding, and agent scenarios — competing directly with DeepSeek, Kimi, and GPT-class frontier models. Full benchmark comparison on our Hy4 vs DeepSeek page.
| Model | Type | Positioning |
|---|---|---|
| Hy4 Preview | Tencent Hunyuan expert MoE | Complex reasoning / coding / agents |
| Hy3 Preview | Tencent Hunyuan MoE | General-purpose open-source |
| DeepSeek V-series | MoE | Reasoning / coding |
| Kimi / GPT-class | MoE / dense | General assistant |
Hy4 Preview is currently served as a hosted API — the model ID is
tencent/hy4-preview on OpenRouter. It is OpenAI-compatible, so you can call it
through any gateway that supports the OpenAI API format. You can create a free account at
tokenra.io/register
to start calling the model right away. See API docs for examples.
Hy4 Preview is Tencent Hunyuan's next-generation Mixture-of-Experts flagship model and the successor to Hy3. It has 770B total parameters with 49B activated per token, a 1,048,576 token context window, and a maximum output of 64,000 tokens.
Hy4 uses a Mixture-of-Experts architecture with a 78-layer backbone. The first layer is a standard dense FFN; the remaining 77 layers each contain 256 routed experts plus 1 shared expert, with every token activating the top-8 routed experts along with the shared expert. A native MTP layer (10B total parameters, 0.7B activated) is built in for speculative decoding.
Hy4 Preview is available through OpenAI-compatible API gateways. You can create a free account at tokenra.io/register to start calling the tencent/hy4-preview model ID.
Tencent has publicly documented Hy4's architecture on the official research page and HuggingFace. The release is currently served as a hosted API (model ID tencent/hy4-preview on OpenRouter). For the latest weight availability and license, check the official HuggingFace repository.
On OpenRouter, Hy4 Preview is priced at $0.834 per 1M input tokens and $2.501 per 1M output tokens, with a 1,048,576 token context window.
Hy4 Preview supports a 1,048,576 token context window with a maximum output of 64,000 tokens.