Tencent Hy4 preview arrives on Vega

Tencent released Hy4 preview on August 28, 2026. It is a Mixture-of-Experts model with 770B total parameters and 49B active parameters, and a context window that exceeds 1 million tokens. Tencent built it for real-world productivity work: coding, office tasks, and scientific research.

Hy4 preview is now available on vega.chat, ru.vega.chat, and through the api.vega.chat API. vega.chat runs the model in the chat interface with no subscription required. ru.vega.chat adds it to the unlimited Autopilot mode that covers chat, web, and images. api.vega.chat gives every account a test deposit, free models, and discounts on new releases, with full OpenRouter compatibility.

How it performs

In a blind internal evaluation at Tencent, 163 experts rated the model across 203 engineering tasks. Hy4 preview scored 2.99 out of 4.00 on average, ahead of GLM-5.3 at 2.92 and Kimi K3 at 2.94.

The training data was co-created with Tencent specialists across software engineering, gaming, finance, and security, and tuned through joint work with products like WorkBuddy. That shaped the gains across several real-world scenarios.

Software engineering

Hy4 preview handles long-context development tasks with stronger understanding, planning, debugging, and validation. Front-end output also improved in visual quality and interaction.

Office and analytical work

The model reads complex working environments better and does stronger financial analysis. It is tuned for data analysis and cross-document collaboration, covering the workflow from raw information through to documents, spreadsheets, and presentations.

Game development

A single natural-language request can produce a playable prototype that works with game engines. From there, developers refine the project over multiple turns.

Scientific research

Hy4 preview reasons through and solves harder research problems, with clear gains in AI research and development, molecular dynamics simulation, condensed-matter physics, and fundamental mathematics.

A model that helped train and tune itself

Hy4 preview took part in its own development. It proposed approaches for training methods, data strategies, evaluation frameworks, and low-level operators, then ran experiments and iterated on the results. The code, logs, and feedback fed into the next round. Tencent calls this an early recursive self-improvement loop.

The model also analyzed bottlenecks in its inference system and optimized operator fusion and communication. End-to-end throughput rose 31.8% over the baseline, with consistent gains across different context lengths and concurrency levels.

Pricing

Tencent's listed API pricing is $0.834 per million input tokens, $2.501 per million output tokens, and $0.042 per million tokens on cache hits.

What's next

Hy4 preview is the first release in the Hy4 series. Tencent says the next batch of models is coming soon, and that real-world feedback from this preview feeds back into development.

Try Hy4 preview on Vega

Pick Hy4 preview in the model catalog on vega.chat or ru.vega.chat, or call it through api.vega.chat. Give it a long-context coding task, a financial analysis, or a research problem, and compare the result with other flagships in the same interface.