Hy4 Preview is Tencent's new-generation flagship model, part of the Hunyuan Hy4 family. It is a 770-billion-parameter Mixture-of-Experts model with 49 billion active parameters, trained alongside Tencent's internal software engineering, game development, finance, and security teams around the tasks those teams actually ship, with a native 10-billion-parameter MTP layer built in for speculative decoding.
| Developer | Tencent |
|---|---|
| Model ID | tencent/hy4-preview |
| Release Date | August 28, 2026 |
| Context Window | 1,048,576 tokens (~1M) |
| Max Output Tokens | 64,000 |
| Parameters | 770B total / 49B active (MoE) |
| Modalities | Text (input) → Text (output) |
| Capabilities | Reasoning, Streaming, Structured output, Tool calling |
Large-scale MoE with efficient activation
78 layers — one dense FFN layer plus 77 MoE layers, each with 256 routed experts and 1 shared expert, top-8 routing per token. A native 10B-parameter MTP layer (0.7B activated) handles speculative decoding for faster inference.
Built around real engineering work
Trained with Tencent's internal software engineering, game development, finance, and security teams on the tasks those teams actually ship — including front-end visual and interaction quality, not just benchmark scores.
Open weights, Apache 2.0 license
Both Hy4 Preview and an FP8 variant are available for self-hosted deployment, with dedicated tool-calling and reasoning parsers for agent frameworks.
Long-horizon software engineering
Understanding, planning, debugging, and verifying extended development tasks across a 1M-token context window large enough to hold full codebases or long agent sessions.
Tool-calling agent pipelines
Dedicated tool-call and reasoning parsers make it a practical backbone for agent frameworks that need reliable, structured tool use.
Self-hosted deployments
Open weights under Apache 2.0, including an FP8 variant, for teams that need to run the model on their own infrastructure.
Measured on the AI/ML API gateway across live traffic. Figures are medians unless stated. Last updated 2026-10-12.
Software engineering: Hy4 preview is documented as suited for understanding, planning, debugging, and verifying long-horizon development tasks, with improved front-end visual taste and interaction quality.
Office and analysis: It is documented as suited for turning messy information spread across many files into documents, spreadsheets, and presentations, including data analysis, equations, and financial models.
Game development: It is documented as suited for turning a single prompt into a playable prototype and working fluently with game engines across multiple turns.
Scientific research: It is documented as suited for harder research questions, with strengths noted in AI research, molecular dynamics, condensed matter physics, and pure mathematics.
import requests r = requests.post( "https://api.aimlapi.com/v1/chat/completions", headers={"Authorization": "Bearer " + AIMLAPI_KEY}, json={ "model": "tencent/hy4-preview", "messages": [ { "role": "user", "content": "Hello!" } ] }, ) print(r.json())
const r = await fetch("https://api.aimlapi.com/v1/chat/completions", {
method: "POST",
headers: {
Authorization: `Bearer ${process.env.AIMLAPI_KEY}`,
"Content-Type": "application/json",
},
body: JSON.stringify({
"model": "tencent/hy4-preview",
"messages": [
{
"role": "user",
"content": "Hello!"
}
]
}),
});
console.log(await r.json());
curl -X POST https://api.aimlapi.com/v1/chat/completions \ -H "Authorization: Bearer $AIMLAPI_KEY" \ -H "Content-Type: application/json" \ -d '{"model":"tencent/hy4-preview","messages":[{"role":"user","content":"Hello!"}]}'
OpenAI-compatible — swap the base URL and it works with your existing SDK.
| Type | Price |
|---|---|
| Input | |
| Output | |
| Cached input |
| Benchmark | Score | What it measures | Source | Retrieved |
|---|---|---|---|---|
| GPQA Diamond | 92.3 | Google-proof graduate science questions (hardest subset) | Source | September 21, 2026 |
| Model | Input | Output | Context | Best for |
|---|---|---|---|---|
Hy4 Preview This page | Long-horizon coding + agentic works | |||
| Reasoning + agents | ||||
| Long-context, multimodal & agentic workflows | ||||
| Reasoning + agents |
Hy4 Preview has a 1,048,576 tokens context window and can return up to 64,000 tokens.
Hy4 Preview takes text as input and returns text.
Use tencent/hy4-preview as the model id. Requests go to https://api.aimlapi.com/v1/chat/completions.
Hy4 Preview became available on August 28, 2026.
Hy4 Preview is priced at input $1.1470836 / 1M tokens, output $3.4398754 / 1M tokens, cached input $0.0577668 / 1M tokens.
Yes, Hy4 Preview can stream responses as they are generated.
Hy4 Preview was built by Tencent.
Yes, Hy4 Preview supports function calling, allowing it to invoke external tools during a conversation.
Hy4 Preview is a mixture-of-experts model with 49B active parameters out of 770B total.
