Skip to main content
ToolPotion
AI Advisor
Submit a Tool (Free)

Hy4 Preview

No ratings yet
LLMs#34AI Models & LLMsOpen source · Web

Hy4 Preview is a new-generation Mixture-of-Experts model developed by the Tencent Hy Team. It features 770 billion parameters and is designed for productivity across various domains, including software engineering, game development, and scientific research.

Visit Website
Share
Hy4 Preview screenshot
Hy4 Preview screenshot

Description

Hy4 Preview is a cutting-edge Mixture-of-Experts (MoE) model developed by the Tencent Hy Team. It boasts a total of 770 billion parameters, with 49 billion activated per token. The model's architecture consists of 78 layers, utilizing a combination of dense FFN and MoE layers, each containing 256 routed experts and one shared expert. This design allows for the activation of the top-8 routed experts per token, enhancing its processing capabilities.

Read More

The model employs Gated DeepSeek Sparse Attention (Gated DSA) with IndexCache for efficient cross-layer sparse index reuse. Its residual pathway uses identity Hyper-Connections (iHC) to improve inter-layer information flow. Hy4 Preview is scaled in terms of model size, context length, and training data, making it a leader in open-source AI models.

Hy4 Preview is designed to excel in various domains, including software engineering, where it aids in understanding, planning, debugging, and verifying long-horizon tasks. In game development, it can transform a single prompt into a playable prototype, working seamlessly with game engines. For office and analysis tasks, it converts messy context into shareable artifacts like documents and spreadsheets. In scientific research, it offers strong reasoning and problem-solving capabilities.

The model is open-source and available on platforms like Hugging Face, ModelScope, GitCode, and CNB. It supports deployment with vLLM or SGLang and offers a comprehensive model finetuning pipeline. Hy4 Preview is released under the Apache License 2.0, ensuring its accessibility and adaptability for various applications.

  • CategoryAI Models & LLMs
  • Pricing ModelOpen source
  • DeploymentCloud
  • PlatformsWeb

Hy4 Preview Highlights

  • Mixture-of-Experts Architecture

  • 770 Billion Parameters

  • Gated DeepSeek Sparse Attention

  • Identity Hyper-Connections

  • Open Source

  • Fine-tuning Support

  • Speculative Decoding

  • Available on Hugging Face

Getting Started with Hy4 Preview

  1. Access model page: Visit the official Hugging Face page

  2. Authenticate: Log in or create an account

  3. Configure: Set up model parameters and environment

  4. Send prompts: Interact with the model using API calls

  5. Fine-tune: Customize the model for specific tasks

Hy4 Preview's Use Cases

  • Software Engineering
  • Game Development
  • Office Analysis
  • Scientific Research
  • AI Research

FAQ from Hy4 Preview

From Tencent

Hy4 Preview Reviews

Hy4 Preview Reviews

Loading...

Popular AI Tools Like Hy4 Preview

Hy3

LLMs

Hy3 is a 295B-parameter Mixture-of-Experts model developed by the Tencent Hy Team. It excels in reasoning, agentic, and long-context tasks, offering reliable performance across…

AI Models & LLMs

Ranked #24 in LLMs

View Tool

Hy3

LLMs

295B-parameter Mixture-of-Experts model developed by the Tencent Hy Team. It excels in reasoning, agentic, and long-context tasks, offering reliable performance across…Hy3 is a 295B-parameter Mixture-of-Experts model developed by the Tencent Hy Team. It excels in reasoning, agentic, and long-context tasks, offering reliable performance across…

AI Models & LLMs
Ranked #24 in LLMs

Laguna M.1

LLMs

Laguna M.1 is a 225B parameter Mixture-of-Experts model designed for agentic coding and long-horizon tasks. It features high-capacity expert routing and global attention…

AI Models & LLMs

Ranked #35 in LLMs

View Tool

Laguna M.1

LLMs

225B parameter Mixture-of-Experts model designed for agentic coding and long-horizon tasks. It features high-capacity expert routing and global attention…Laguna M.1 is a 225B parameter Mixture-of-Experts model designed for agentic coding and long-horizon tasks. It features high-capacity expert routing and global attention…

AI Models & LLMs
Ranked #35 in LLMs

DeepSeek-V4-Pro

LLMs

DeepSeek-V4-Pro is an advanced AI model designed for efficient million-token context intelligence. It features a hybrid attention architecture and is pre-trained on over 32…

AI Models & LLMs
View Tool

DeepSeek-V4-Pro

LLMs

Advanced AI model designed for efficient million-token context intelligence. It features a hybrid attention architecture and is pre-trained on over 32…DeepSeek-V4-Pro is an advanced AI model designed for efficient million-token context intelligence. It features a hybrid attention architecture and is pre-trained on over 32…

FeaturedAI Models & LLMs
See the Top 50 LLMs

DeepSeek

LLMs

DeepSeek is an AI research company dedicated to developing general artificial intelligence. It offers open-source frontier LLMs like DeepSeek-V4, DeepSeek-R1, and DeepSeek-Coder,…

AI Models & LLMs

Ranked #2 in LLMs

View Tool

DeepSeek

LLMs

AI research company dedicated to developing general artificial intelligence. It offers open-source frontier LLMs like DeepSeek-V4, DeepSeek-R1, and DeepSeek-Coder,…DeepSeek is an AI research company dedicated to developing general artificial intelligence. It offers open-source frontier LLMs like DeepSeek-V4, DeepSeek-R1, and DeepSeek-Coder,…

FeaturedAI Models & LLMs
Ranked #2 in LLMs

Kimi-K2-Instruct Model

LLMs

Kimi-K2-Instruct is a state-of-the-art mixture-of-experts language model designed for advanced AI tasks. It excels in reasoning, coding, and tool use, offering robust performance…

AI Models & LLMs

Ranked #44 in LLMs

View Tool

Kimi-K2-Instruct Model

LLMs

State-of-the-art mixture-of-experts language model designed for advanced AI tasks. It excels in reasoning, coding, and tool use, offering robust performance…Kimi-K2-Instruct is a state-of-the-art mixture-of-experts language model designed for advanced AI tasks. It excels in reasoning, coding, and tool use, offering robust performance…

AI Models & LLMs
Ranked #44 in LLMs

DeepSeek-V4-Pro-0813

LLMs

DeepSeek-V4-Pro-0813 is an advanced language model designed to enhance agentic capabilities and performance in production environments. It builds upon the DeepSeek-V4-Pro…

AI Models & LLMs

Ranked #6 in LLMs

View Tool

DeepSeek-V4-Pro-0813

LLMs

Advanced language model designed to enhance agentic capabilities and performance in production environments. It builds upon the DeepSeek-V4-Pro…DeepSeek-V4-Pro-0813 is an advanced language model designed to enhance agentic capabilities and performance in production environments. It builds upon the DeepSeek-V4-Pro…

AI Models & LLMs
Ranked #6 in LLMs

DeepSeek-V4-Pro-DSpark

LLMs

DeepSeek-V4-Pro-DSpark is an advanced language model featuring a hybrid attention architecture and manifold-constrained hyper-connections. It supports a context length of one…

AI Models & LLMs
View Tool

DeepSeek-V4-Pro-DSpark

LLMs

Advanced language model featuring a hybrid attention architecture and manifold-constrained hyper-connections. It supports a context length of one…DeepSeek-V4-Pro-DSpark is an advanced language model featuring a hybrid attention architecture and manifold-constrained hyper-connections. It supports a context length of one…

AI Models & LLMs
See the Top 50 LLMs

DeepSeek-V4-Flash-0731

LLMs

DeepSeek-V4-Flash-0731 is an advanced language model that enhances agentic capabilities. It outperforms previous versions and is competitive with top proprietary models, offering…

AI Models & LLMs

Ranked #16 in LLMs

View Tool

DeepSeek-V4-Flash-0731

LLMs

Advanced language model that enhances agentic capabilities. It outperforms previous versions and is competitive with top proprietary models, offering…DeepSeek-V4-Flash-0731 is an advanced language model that enhances agentic capabilities. It outperforms previous versions and is competitive with top proprietary models, offering…

AI Models & LLMs
Ranked #16 in LLMs
1 of 8