Technology
November 16, 2026 1059 words

Tencent Hy4 and Hy4 flash: Computer Use and Ultra-Low-Cost AI

Hams AI Tech
Professional Content Team
Expert in AI Prompts & Professional Tools

Tencent made two significant AI releases in November 2026. On November 4 it launched Hy4, the full version of the flagship it previewed in August and its first model with visual understanding and computer use. Then on November 16 it released Hy4 flash, its lowest-cost model so far, priced at a level that makes high-volume AI almost free. Together they show how Chinese labs are pushing both capability and price at the same time.

Hy4: Tencent's Flagship Learns to Use Computers

Hy4 is the first model in the Hy family that can see and operate software. Tencent says it can work across both command-line and graphical applications, and it was trained on professional software including Blender and CAD tools. That last point is unusual and important for engineers, architects and designers, whose daily work happens inside specialized programs rather than web browsers.

Tencent reports gains over Hy4 Preview in several areas:

    • Instruction following.
    • Long-context understanding across a one million token window.
    • Long-horizon agent tasks and deep research.
    • Front-end and game development.
    • Finance work.

Hy4 is available by API on Tencent Cloud TokenHub and OpenRouter, and inside Tencent's WorkBuddy, CodeBuddy, Yuanbao and ima apps, at the same price as Hy4 Preview. Tencent has said it plans to release Hy4 weights with a system card; at launch, only the Hy4 Preview weights were open.

Hy4 flash: Frontier Features at a Fraction of the Price

Hy4 flash is built for volume. It has 224 billion total parameters, including 64 billion Engram parameters, with 12 billion active per token. It is natively multimodal and supports a one million token context.

The pricing is remarkable:

    • 0.07 dollars per million input tokens.
    • 0.281 dollars per million output tokens.
    • 0.003 dollars per million cached input tokens.

To put that in context, a firm could process millions of words of documents for a few dollars. Tencent says that in an internal blind test with 85 WorkBuddy users, Hy4 flash scored close to Claude Haiku 5.5 on satisfaction, and on its WorkIndex benchmark it costs roughly one twentieth as much per task as Hy4.

Hy4 flash is available in Tencent Cloud TokenHub with 50% off through November 30, free in WorkBuddy through November 30, and in Yuanbao, WeCom, QQ and other Tencent products. OpenRouter availability and open weights are planned.

Why Ultra-Cheap Models Matter

When AI costs almost nothing per request, new workflows become economical. Tasks that were too expensive to automate at scale suddenly make sense:

    • Classifying every incoming email or document in a large organization.
    • Summarizing every customer call rather than a sample.
    • Extracting fields from thousands of invoices or forms.
    • Running quality checks on every draft before human review.

Cheap models are also ideal as the first stage in multi-model pipelines, where they triage work and pass only difficult items to expensive models. Our guide to AI decision models explores a related category built specifically for routing and classification.

Professional Use Cases

Accountants

High-volume transaction categorization and document extraction are natural fits for Hy4 flash. Accountants who want structured workflows can look at the multi-entity consolidation agent for accountants and audit preparation resources.

Engineers and Designers

Hy4's training on Blender and CAD makes it interesting for design automation. Engineers can explore architecture AI resources and architectural plans resources for related prompts.

Marketing and Social Media

Cheap models can generate many content variations for testing. Social media teams may benefit from the viral trend detection agent and social media tools.

Developers

CodeBuddy integration and front-end skills make Hy4 useful for web projects. See programming resources.

Things to Check Before Using Tencent Models

    • Where your data is processed and stored, especially for client or patient information.
    • Whether your clients or regulators restrict cross-border data transfer.
    • Promotional pricing end dates, so budgets do not jump unexpectedly in December.
    • Independent benchmark results, since Tencent's comparisons are internal.

Designing a Two-Tier Workflow

The most effective way to use a model like Hy4 flash is as the first layer of a two-tier workflow. In the first tier, the cheap model reads every item, whether that is an email, an invoice or a support ticket, and labels it: routine, needs attention or escalate. Routine items get a drafted response or an extracted record. Items marked for escalation go to a stronger model or directly to a person. Because the first tier costs fractions of a cent per item, you can afford to process everything rather than a sample, which means fewer things slip through the cracks.

This design also improves human focus. Instead of reading hundreds of routine messages, professionals spend their time on the small number that actually need judgment. Automation platforms such as n8n make this pattern straightforward to build, and our n8n workflow templates include examples you can adapt.

The Bigger Trend

Hy4 flash launched on the same day as Alibaba's Qwen4-Max, and in the same week Meta priced Muse Ember at 0.10 dollars per million input tokens. The cost of capable AI is falling quickly. For professionals, the competitive advantage is shifting from access to models toward knowing what to ask them, which is why proven prompts and workflows matter more than ever.

Frequently Asked Questions

Is Hy4 flash as good as premium models?

Not for the hardest tasks. Tencent compares it with Claude Haiku 5.5, a fast tier, not a flagship. It is excellent for volume work and first-pass drafting.

Can Hy4 operate desktop software?

Yes. Hy4, the flagship, supports computer use across command-line and graphical applications, including professional tools such as Blender and CAD software.

Are the weights open?

Hy4 Preview weights are open. Tencent plans to release Hy4 weights with a system card, and open weights for Hy4 flash are also planned.

What is the cheapest way to try Hy4 flash?

WorkBuddy offers it free through November 30, 2026, and Tencent Cloud TokenHub offers 50% off over the same period.

The Bottom Line

Tencent now offers both a capable computer-use flagship and one of the cheapest multimodal models on the market. Use Hy4 for complex software-driven work and Hy4 flash for high-volume tasks. Explore our AI automation resources and AI agent workflows to put these economics to work.

Related AI News

Topics & Keywords
Tencent Hy4 Hy4 flash Tencent AI WorkBuddy computer use cheap AI model