🚀 Unveiling the Tencent Hy4 Preview

Date: August 28, 2026
Category: Corporate Innovation

Tencent has officially announced the release and open-sourcing of the Hy4 preview. This next-generation large language model (LLM) is specifically engineered to excel in professional productivity environments, offering top-tier performance in scientific research, office administration, and software engineering.

🛠️ Technical Architecture & Specifications

The Hy4 preview represents a massive leap in scale and capability. By expanding the volume of training data, increasing model size, and refining both pre- and post-training phases, Tencent has pushed this model into the elite bracket of open-source AI.

Feature	Specification
Total Parameters	$770\text{B}$
Active Parameters	$49\text{B}$
Context Window	$> 1,000,000$ tokens
Primary Focus	Real-world productivity & professional workflows

🌐 Access and Availability

Users can engage with the model through various channels. To encourage adoption, Tencent is offering several promotional windows:

WorkBuddy & CodeBuddy: Free access for the first two weeks.
Hy3 Legacy Access: Free availability extended until September 30.
Tencent Ecosystem: Integrated into Yuanbao and ima.
Developer APIs: Available via Tencent Cloud TokenHub and OpenRouter.

📊 Performance Benchmarks

Tencent conducted a rigorous internal blind evaluation to test the model's efficacy. The study involved 163 experts assessing 203 distinct engineering tasks.

Expert Scoring (Out of 4.00):

Hy4 preview: 2.99
Kimi K3: 2.94
GLM-5.3: 2.92

🎯 Domain-Specific Capabilities

The model was trained using high-fidelity data curated by specialists in security, finance, gaming, and software engineering.

💻 Software Engineering

The Hy4 preview enhances the development lifecycle by improving:

Long-context management: Better planning, debugging, and validation.
Front-end Development: Superior interaction design and visual quality.

📁 Office Productivity

It streamlines complex corporate workflows, including:

Financial Analysis: Deep understanding of intricate fiscal environments.
Cross-Document Synergy: A seamless pipeline from raw data $\rightarrow$ spreadsheets $\rightarrow$ presentations $\rightarrow$ final documents.

🎮 Game Development

The model can bridge the gap between concept and execution:

Rapid Prototyping: Generates playable prototypes from a single natural-language prompt.
Engine Integration: Works directly with game engines.
Iterative Design: Refines complex projects through multi-turn dialogue.

🔬 Scientific Research

Significant breakthroughs were noted in:

Fundamental Mathematics
Condensed-matter Physics
Molecular Dynamics Simulation
AI R&D

🔄 The Recursive Self-Improvement Loop

In a landmark achievement, Hy4 preview actually assisted in its own creation. ~~Traditional manual tuning~~ was replaced by an automated recursive loop.

"The model proposed new approaches, executed experiments, and iterated based on the resulting logs and feedback, feeding this data back into the next round of exploration."

Self-Optimization Workflow:

Furthermore, the model autonomously identified bottlenecks in its own inference system. By optimizing communication and operator fusion, it achieved a throughput increase of: $\text{Throughput Gain} = +31.8\%$

💰 Pricing Structure

Tencent aims to keep advanced AI accessible through a competitive pricing model:

Token Type	Price (per Million Tokens)
Input	USD $0.834$
Output	USD $2.501$
Cache Hits	USD $0.042$

📅 Future Outlook

Tencent is utilizing a "preview-first" strategy, allowing real-world user feedback to dictate the evolution of the Hunyuan series.

What's next?

Integration of current feedback into the core R&D.
The imminent rollout of additional models within the Hy4 family.

Related Corporate Milestones:

July 30, 2026: Tencent enters the Fortune Global 500 Top 100.
July 25, 2026: AI used to preserve UNESCO heritage in Jingdezhen.
July 20, 2026: Launch of ADP 4.0 Embodied Intelligence.
July 6, 2026: Official release of Hy3.

Example API Implementation:

import tencent_cloud_hub

client = tencent_cloud_hub.Client(api_key="YOUR_KEY")
response = client.chat(
    model="hy4-preview",
    messages=[{"role": "user", "content": "Optimize this molecular simulation code."}]
)
print(response.text)

WeChat QR Code