Total runs:
180.9K
Run Growth:
-191.9K
Growth Rate:
-106.08%
Sponsored by
i10X
tencent / Hy3-preview-Base
Model's Last Updated: April 23 2026
text-generation
Introduction of Hy3-preview-Base
Model Details of Hy3-preview-Base
中文 ｜ English
🖥️ Official Website | 💬 GitHub
Table of Contents
- Model Introduction
- Highlights
- Benchmark Results
- News
- Model Links
- Quickstart
- Deployment
- 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.
| Property | Value |
|---|---|
| Architecture | Mixture-of-Experts (MoE) |
| Total Parameters | 295B |
| Activated Parameters | 21B |
| MTP Layer Parameters | 3.8B |
| Number of Layers (excluding MTP layer) | 80 |
| Number of MTP Layers | 1 |
| Attention Heads | 64 (GQA, 8 KV heads, head dim 128) |
| Hidden Size | 4096 |
| Intermediate Size | 13312 |
| Context Length | 256K |
| Vocabulary Size | 120832 |
| Number of Experts | 192 experts, top-8 activated |
| Supported Precisions | BF16 |
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
| Category | Benchmark (Metric) | # Shots | Kimi-K2 BASE | DeepSeek-V3 BASE | GLM-4.5 BASE | Hy3 preview-Base |
|---|---|---|---|---|---|---|
| #ActivatedParams | - | 32B | 37B | 32B | 21B | |
| #TotalParams | - | 1043B | 671B | 355B | 295B | |
| English | MMLU | 5-shot | 88.24 | 87.68 | 87.73 | 87.42 |
| MMLU-Pro | 5-shot | 65.98 | 63.98 | 63.67 | 65.76 | |
| MMLU-Redux | 5-shot | 87.18 | 86.81 | 86.56 | 86.86 | |
| ARC-Challenge | 0-shot | 96.66 | 94.65 | 96.32 | 95.99 | |
| DROP | 5-shot | 86.40 | 86.50 | 82.90 | 85.50 | |
| PIQA | 4-shot | 84.93 | 84.22 | 84.71 | 84.39 | |
| SuperGPQA | 5-shot | 51.10 | 46.17 | 49.64 | 51.60 | |
| SimpleQA | 5-shot | 34.37 | 26.15 | 29.26 | 26.47 | |
| Code | MBPP-plus | 3-shot | 81.35 | 75.47 | 78.05 | 78.71 |
| CRUXEval-I | 3-shot | 68.01 | 67.79 | 68.51 | 71.19 | |
| CRUXEval-O | 3-shot | 69.62 | 71.00 | 67.75 | 68.38 | |
| LiveCodeBench-v6 | 1-shot | 30.86 | 29.31 | 27.43 | 34.86 | |
| Math | GSM8K | 4-shot | 93.46 | 88.15 | 90.06 | 95.37 |
| MATH | 4-shot | 71.20 | 59.37 | 61.00 | 76.28 | |
| CMath | 4-shot | 90.83 | 85.50 | 89.33 | 91.17 | |
| Chinese | C-Eval | 5-shot | 91.51 | 90.35 | 85.84 | 89.80 |
| CMMLU | 5-shot | 90.72 | 87.90 | 86.46 | 89.61 | |
| Chinese-simpleQA | 5-shot | 74.58 | 68.72 | 68.49 | 69.73 | |
| Multilingual | MMMLU | 5-shot | 77.63 | 79.54 | 79.26 | 80.15 |
| INCLUDE | 5-shot | 75.66 | 77.86 | 76.27 | 78.64 |
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.
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).
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.
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.
News
- [2026-04-23] 🔥 We open-source Hy3 preview model weights on Hugging Face , ModelScope , and GitCode .
Model Links
| Model Name | Description | Hugging Face | ModelScope | GitCode |
|---|---|---|---|---|
| Hy3 preview | Instruct model | 🤗 Model | Model | Model |
| Hy3 preview-Base | Pre-trained base model | 🤗 Model | Model | Model |
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-Base",
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_effortto"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-Base \
--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-Base
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-Base \
--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-Base
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:
Hy3 preview is developed by the Tencent Hy Team.
Runs of tencent Hy3-preview-Base on huggingface.co
485
Total runs
45
24-hour runs
480
3-day runs
480
7-day runs
480
30-day runs
More Information About Hy3-preview-Base huggingface.co Model
Url of Hy3-preview-Base
Hy3-preview-Base huggingface.co Url
Provider of Hy3-preview-Base huggingface.co
Other API from tencent
Updated:January 13 2026
Total runs:
90.7K
Run Growth:
-582.4K
Growth Rate:
-642.43%
Total runs:
74.9K
Run Growth:
-22.8K
Growth Rate:
-30.29%
Total runs:
40.9K
Run Growth:
9.9K
Growth Rate:
24.14%
Total runs:
27.0K
Run Growth:
23.1K
Growth Rate:
85.39%
Total runs:
26.7K
Run Growth:
-88.0K
Growth Rate:
-329.66%
Total runs:
23.0K
Run Growth:
-72.5K
Growth Rate:
-350.46%
Total runs:
20.8K
Run Growth:
-1.7K
Growth Rate:
-8.22%
Total runs:
16.9K
Run Growth:
8.8K
Growth Rate:
52.28%
Total runs:
16.7K
Run Growth:
3.9K
Growth Rate:
24.68%
Total runs:
16.4K
Run Growth:
-6.2K
Growth Rate:
-37.91%
Total runs:
13.5K
Run Growth:
4.0K
Growth Rate:
30.91%
Total runs:
8.8K
Run Growth:
2.4K
Growth Rate:
27.37%
Total runs:
6.4K
Run Growth:
-1.3K
Growth Rate:
-19.58%
Total runs:
5.2K
Run Growth:
1.5K
Growth Rate:
28.52%
Total runs:
4.3K
Run Growth:
-1.4K
Growth Rate:
-32.89%
Total runs:
3.3K
Run Growth:
3.2K
Growth Rate:
95.09%
Total runs:
2.9K
Run Growth:
626
Growth Rate:
21.80%
Total runs:
2.1K
Run Growth:
972
Growth Rate:
46.11%
Total runs:
2.0K
Run Growth:
1.4K
Growth Rate:
59.02%
Total runs:
1.9K
Run Growth:
1.4K
Growth Rate:
72.70%
Total runs:
1.9K
Run Growth:
709
Growth Rate:
36.57%
Total runs:
1.6K
Run Growth:
1.2K
Growth Rate:
71.70%
Total runs:
1.6K
Run Growth:
-4.0K
Growth Rate:
-242.98%
Total runs:
1.6K
Run Growth:
853
Growth Rate:
54.68%
Total runs:
1.4K
Run Growth:
-155
Growth Rate:
-10.76%
Total runs:
1.0K
Run Growth:
332
Growth Rate:
31.68%
Total runs:
1.0K
Run Growth:
150
Growth Rate:
14.90%
Total runs:
926
Run Growth:
-1.8K
Growth Rate:
-193.95%
Total runs:
851
Run Growth:
369
Growth Rate:
43.36%
Total runs:
804
Run Growth:
-99
Growth Rate:
-12.31%
Total runs:
801
Run Growth:
-343
Growth Rate:
-42.82%
Total runs:
693
Run Growth:
-114
Growth Rate:
-16.45%
Total runs:
613
Run Growth:
353
Growth Rate:
57.59%
Total runs:
574
Run Growth:
-100
Growth Rate:
-17.42%
Total runs:
524
Run Growth:
-1.5K
Growth Rate:
-294.47%
Total runs:
471
Run Growth:
-576
Growth Rate:
-139.81%
Total runs:
396
Run Growth:
-106
Growth Rate:
-26.77%
Total runs:
327
Run Growth:
56
Growth Rate:
17.13%
Total runs:
294
Run Growth:
252
Growth Rate:
85.71%
Total runs:
282
Run Growth:
249
Growth Rate:
88.30%
Total runs:
280
Run Growth:
-137
Growth Rate:
-48.93%
Total runs:
277
Run Growth:
-69
Growth Rate:
-24.91%
Total runs:
249
Run Growth:
190
Growth Rate:
76.31%
Total runs:
234
Run Growth:
-105
Growth Rate:
-44.87%
Total runs:
208
Run Growth:
-42
Growth Rate:
-20.19%
Total runs:
184
Run Growth:
137
Growth Rate:
74.46%
Total runs:
171
Run Growth:
129
Growth Rate:
75.44%
Total runs:
149
Run Growth:
-84
Growth Rate:
-56.38%
Total runs:
143
Run Growth:
90
Growth Rate:
62.94%
Total runs:
139
Run Growth:
122
Growth Rate:
87.77%
Total runs:
121
Run Growth:
121
Growth Rate:
100.00%
Total runs:
98
Run Growth:
64
Growth Rate:
65.31%
Total runs:
95
Run Growth:
63
Growth Rate:
66.32%
Total runs:
88
Run Growth:
-57
Growth Rate:
-64.77%
Total runs:
85
Run Growth:
39
Growth Rate:
45.88%
Total runs:
85
Run Growth:
25
Growth Rate:
29.41%
Total runs:
54
Run Growth:
-38
Growth Rate:
-70.37%
Total runs:
53
Run Growth:
3
Growth Rate:
5.66%
Total runs:
49
Run Growth:
-205
Growth Rate:
-418.37%
Total runs:
47
Run Growth:
-45
Growth Rate:
-95.74%
Total runs:
38
Run Growth:
6
Growth Rate:
15.79%
Total runs:
36
Run Growth:
-1
Growth Rate:
-2.78%
Total runs:
35
Run Growth:
4
Growth Rate:
11.43%
Total runs:
35
Run Growth:
-2
Growth Rate:
-5.88%
Total runs:
28
Run Growth:
26
Growth Rate:
92.86%
Total runs:
27
Run Growth:
10
Growth Rate:
37.04%
Total runs:
23
Run Growth:
-13
Growth Rate:
-56.52%
Total runs:
18
Run Growth:
7
Growth Rate:
38.89%
Total runs:
14
Run Growth:
-6
Growth Rate:
-42.86%
Total runs:
8
Run Growth:
-20
Growth Rate:
-250.00%