Hy3-preview

<p align="left"> <a href="https://huggingface.co/tencent/Hy3-preview/blob/main/README_CN.md">中文</a>&nbsp;｜&nbsp;English </p> <br> <p align="center"> <img src="assets/logo-en.png" width="400"/> <br> </p> <div align="center" style="line-height: 1;">

           

</div> <p align="center"> 🖥️&nbsp;<a href="https://aistudio.tencent.com/"><b>Official Website</b></a>&nbsp;&nbsp;|&nbsp;&nbsp; 💬&nbsp;<a href="https://github.com/Tencent-Hunyuan/Hy3-preview"><b>GitHub</b></a></p>

---

Table of Contents

• Model Introduction
• Highlights
• Benchmark Results
  • STEM & Reasoning
  • Context Learning & Instruction Following
  • Code & Agent
• News
• Model Links
• Quickstart
• Deployment
  • vLLM
  • SGLang
• Training
• Quantization
• License
• Contact Us

---

Model Introduction

Hy3 preview is a 295B-parameter Mixture-of-Experts (MoE) model with 21B active parameters and 3.8B MTP layer parameters, developed by the Tencent Hy Team. Hy3 preview is the first model trained on our rebuilt infrastructure, and the strongest we've shipped so far. It improves significantly on complex reasoning, instruction following, context learning, coding, and agent tasks.

Highlights

• STEM & Reasoning — Complex reasoning underpins everything else. Hy3 preview performs well on challenging STEM benchmarks like FrontierScience-Olympiad and IMOAnswerBench, and achieved excellent results in the Tsinghua Qiuzhen College Math PhD qualifying exam (Spring '26) and the China High School Biology Olympiad (CHSBO 2025), demonstrating generalizable reasoning capacity.

• Context Learning & Instruction Following — Real-world tasks require the ability to parse messy, lengthy contexts and follow complex rules. We built CL-bench and CL-bench-Life from our own business scenarios to innovatively measure context learning ability. Hy3 preview exhibits solid gains in both context learning and instruction following capabilities.

• Code & Agent — Coding and agents saw the biggest gains. With a rebuilt RL infrastructure and larger-scale training tasks, we posted competitive scores across mainstream coding agent benchmarks (SWE-bench Verified, Terminal-Bench 2.0) and search agent benchmarks (BrowseComp, WideSearch).

Benchmark Results

Pre-trained Model Performance

Instruct Model Performance

STEM & Reasoning

Complex reasoning underpins everything else. Hy3 preview performs well on challenging STEM benchmarks like FrontierScience-Olympiad and IMOAnswerBench. It also achieved excellent results in the Tsinghua Qiuzhen College Math PhD qualifying exam (Spring '26) and the China High School Biology Olympiad (CHSBO 2025), demonstrating a high degree of generalizable reasoning capacity.

<p align="center"><img src="assets/bench_stem.jpg" width="800" alt="STEM & Reasoning benchmarks"/></p>

Context Learning & Instruction Following

Real-world tasks require the ability to parse messy, lengthy contexts and follow complex rules. We built CL-bench and CL-bench-Life from our own business scenarios to innovatively measure context learning ability. Hy3 preview exhibits solid gains in both context learning and instruction following capabilities.

<p align="center"><img src="assets/bench_context.jpg" width="800" alt="Context Learning & Instruction Following benchmarks"/></p>

Code & Agent

Coding and agents saw the biggest gains. With a rebuilt RL infrastructure and larger-scale training tasks, we posted competitive scores across mainstream coding agent benchmarks (SWE-bench Verified, Terminal-Bench 2.0) and search agent benchmarks (BrowseComp, WideSearch).

<p align="center"><img src="assets/bench_agent_overview_v3.jpg" width="800" alt="Agent benchmarks overview"/></p>

Coding is about whether a model can execute in a development environment. Search is about whether it can find and combine information from the open web. Both matter for complex agent scenarios like OpenClaw. Hy3 preview scores well on ClawEval and WildClawBench — a sign that its agent capabilities are becoming practical.

<p align="center"><img src="assets/bench_claw_agent.png" width="800" alt="Claw Agent benchmarks"/></p>

Beyond public benchmarks, we built internal evaluation sets to test the model in real development scenarios. On Hy-Backend (backend-focused tasks), Hy-Vibe Bench (real-user dev workflows), and Hy-SWE Max, Hy3 preview scores competitively against other open-source models.

<p align="center"><img src="assets/bench_claw_agent2.jpg" width="800" alt="Internal benchmarks"/></p>

News

• [2026-04-23] 🔥 We open-source Hy3 preview model weights on Hugging Face, ModelScope, and GitCode.

Model Links

Quickstart

Deploy Hy3 preview with vLLM or SGLang first, then call the OpenAI-compatible API:

from openai import OpenAI

client = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY")

response = client.chat.completions.create(
    model="hy3-preview",
    messages=[
        {"role": "user", "content": "Hello! Can you briefly introduce yourself?"},
    ],
    temperature=0.9,
    top_p=1.0,
    # reasoning_effort: "no_think" (default, direct response), "low", "high" (deep chain-of-thought)
    extra_body={"chat_template_kwargs": {"reasoning_effort": "no_think"}},
)
print(response.choices[0].message.content)

Recommended parameters: temperature=0.9, top_p=1.0.

Reasoning mode: Set reasoning_effort to "high" for complex tasks (math, coding, reasoning) or "no_think" for direct responses.

See the Deployment section below for how to start the API server.

Deployment

Hy3-preview has 295B parameters in total. To serve it on 8 GPUs, we recommend using H20-3e or other GPUs with larger memory capacity.

vLLM

Build vLLM from source:

uv venv --python 3.12 --seed --managed-python
source .venv/bin/activate
git clone https://github.com/vllm-project/vllm.git
cd vllm
uv pip install --editable . --torch-backend=auto

Start the vLLM server with MTP enabled:

vllm serve tencent/Hy3-preview \
  --tensor-parallel-size 8 \
  --speculative-config.method mtp \
  --speculative-config.num_speculative_tokens 1 \
  --tool-call-parser hy_v3 \
  --reasoning-parser hy_v3 \
  --enable-auto-tool-choice \
  --served-model-name hy3-preview

SGLang

Build SGLang from source:

git clone https://github.com/sgl-project/sglang
cd sglang
pip3 install pip --upgrade
pip3 install "transformers>=5.6.0"
pip3 install -e "python"

Launch SGLang server with MTP enabled:

python3 -m sglang.launch_server \
  --model tencent/Hy3-preview \
  --tp 8 \
  --tool-call-parser hunyuan \
  --reasoning-parser hunyuan \
  --speculative-num-steps 1 \
  --speculative-eagle-topk 1 \
  --speculative-num-draft-tokens 2 \
  --speculative-algorithm EAGLE \
  --served-model-name hy3-preview

Training

Hy3 preview provides a complete model training pipeline, supporting both full fine-tuning and LoRA fine-tuning, with DeepSpeed ZeRO configurations and LLaMA-Factory integration.

For detailed training documentation, please refer to: Training Guide

Quantization

We provide AngelSlim, a more accessible, comprehensive, and efficient toolkit for large model compression. AngelSlim supports a comprehensive suite of compression tools for large-scale multimodal models, including common quantization algorithms, low-bit quantization, and speculative sampling.

License

Hy3 preview is released under the Tencent Hy Community License Agreement. See LICENSE for details.

Contact Us

If you would like to leave a message for our R&D and product teams, welcome to contact us. You can also reach us via email:

📧 [email protected]

---

<p align="center"> <i>Hy3 preview is developed by the Tencent Hy Team.</i> </p>