Overview
Tencent Hy is Tencent's global foundation-model brand, formerly known as Tencent Hunyuan. The lineup spans language, reasoning, agent, image, video, and 3D models. Its current flagship language release is Hy3, officially launched on July 6, 2026 after the April Hy3 Preview.
Hy3 is a hybrid fast-and-slow-thinking model built on a mixture-of-experts architecture with 295 billion total parameters, 21 billion active parameters, and a context window of up to 256K tokens. Tencent says the official release improves stability and cost efficiency over the preview and has already been adopted across WorkBuddy/CodeBuddy, Yuanbao, Marvis, ima, and other Tencent products.
For teams evaluating an AI agent or coding model, the important decision is not only benchmark score. Hy3 combines open-weight availability, Tencent Cloud TokenHub access, long context, and formal Tencent product integration. That makes it most relevant to developers who want a current Chinese frontier model with both hosted and self-hosting paths.
Key Features
295B MoE architecture - Hy3 uses a mixture-of-experts design with 295B total parameters and 21B active parameters. Sparse activation is intended to improve serving efficiency compared with activating an equivalent dense model for every token.
256K-token context window - The model supports long documents, codebases, agent traces, and multi-file reasoning tasks that exceed ordinary chat limits. This is useful for coding agents, long-form analysis, and enterprise knowledge work.
Fast-and-slow-thinking behavior - Tencent describes Hy3 as combining fast and slow thinking in one model. The practical benefit is a single model family that can handle routine responses and more deliberate reasoning without forcing users to switch products.
Tencent ecosystem integration - The official release is integrated across Tencent products including WorkBuddy/CodeBuddy, Yuanbao, Marvis, ima, and others. This gives enterprise users more routes to evaluate Hy3 beyond a standalone model card.
TokenHub API access - Hy3 is available through Tencent Cloud TokenHub. The official Chinese pricing page lists Hy3 at CNY 1 per million input tokens, CNY 4 per million output tokens, and CNY 0.25 per million cached-input tokens.
Open-weight route - The Hy3 line is distributed through Tencent's model channels under Tencent's custom Hy Community License. Organizations should review the license and commercial terms before using weights in revenue-generating products.
How It Compares
Hy3 competes with Chinese and international frontier models used for reasoning, coding, and agent workflows. Its strongest differentiators are the combination of large MoE scale, long context, hosted Tencent Cloud access, and open-weight deployment options.
| Tencent Hy3 | GPT-5.6 Sol | Claude Opus 4.8 | GLM-5 | |
|---|---|---|---|---|
| Primary positioning | Open-weight reasoning and agent model | Flagship OpenAI model | Closed frontier assistant model | Chinese frontier model |
| Parameters | 295B total, 21B active | Undisclosed | Undisclosed | Undisclosed |
| Context | Up to 256K | Up to 1M in published evals | Up to 1M in some surfaces | Up to 1M in some surfaces |
| Hosted API | Tencent Cloud TokenHub | OpenAI API | Anthropic API | Varies by provider |
| Open weights | Yes, custom Tencent license | No | No | Partial/provider-dependent |
The main tradeoff is ecosystem maturity. OpenAI and Anthropic have broader global developer tooling and third-party integrations. Tencent Hy3 may be more attractive when cost, Chinese-language ecosystem alignment, open weights, or Tencent Cloud deployment matter more.
Pricing & Plans
Tencent Hy has multiple access routes: product integrations, TokenHub API access, and model materials for self-hosted evaluation. TokenHub pricing is usage-based rather than a flat subscription.
The official Tencent Cloud Chinese pricing page lists Hy3 at:
| Token type | Official listed price |
|---|---|
| Input tokens | CNY 1 per million tokens |
| Output tokens | CNY 4 per million tokens |
| Cached input | CNY 0.25 per million tokens |
Hy3 Preview remains separately listed with length-tiered pricing. Do not use Preview pricing for the official Hy3 release. If you are buying through Tencent Cloud International, a third-party API platform, or a token plan, verify the current region, currency, quota, and billing unit before estimating production cost.
Best For
- Tencent Cloud users who want a current hosted reasoning model with TokenHub billing.
- Developers building coding agents, search agents, or long-document workflows.
- Research and ML teams evaluating open-weight frontier-class MoE models.
- Enterprises in supported Tencent regions that need Chinese ecosystem alignment.
- Cost-sensitive teams comparing hosted reasoning models across providers.
FAQ
What is Tencent Hy?
Tencent Hy is Tencent's global foundation-model brand, formerly known as Tencent Hunyuan. It covers language, reasoning, agent, image, video, and 3D models.
What is Hy3?
Hy3 is Tencent's official July 2026 language-model release built on a 295B-parameter MoE architecture with 21B active parameters and up to 256K context.
Is Hy3 different from Hy3 Preview?
Yes. Hy3 Preview was the April 2026 preview checkpoint. The official Hy3 release followed on July 6, 2026 with Tencent-positioned improvements in stability, cost efficiency, and product integration.
How much does the Tencent Hy3 API cost?
Tencent Cloud's Chinese TokenHub pricing page lists official Hy3 at CNY 1 per million input tokens, CNY 4 per million output tokens, and CNY 0.25 per million cached-input tokens.
Is Tencent Hy3 open source?
Hy3 is open weight under Tencent's custom Hy Community License rather than an OSI-standard open-source license. Review the license before commercial deployment.
What context length does Hy3 support?
Tencent states Hy3 supports up to 256K tokens of context. That is substantial for code and document workflows, though shorter than some models that advertise 1M-token contexts.
Who should consider Hy3?
Hy3 is most relevant for teams that need long-context reasoning, coding-agent capability, Tencent Cloud access, Chinese ecosystem alignment, or an open-weight alternative to closed API-only models.
