Tencent Hunyuan

What is Tencent Hunyuan?

Tencent Hunyuan is a multi-modal foundational model and generative AI technology suite developed by Tencent; its English name is Hunyuan, while the new version of this platform in 2026 will use the name Hy. It is not a standalone chat application, but rather a family of models that cover text, code, agents, visual content, images, videos, 3D data, voice, and embodied intelligence. Services are provided through Tencent Cloud APIs as well as products such as Yuanbao, ima, WorkBuddy, CodeBuddy, and Tencent Yuanqi.

Ordinary users who want to chat directly, search for information, and work with files mainly use Tencent Yuanbao; developers, on the other hand, use Tencent Cloud TokenHub or the Hunyuan API to access these models.

To create an agent, it is possible to use Tencent Yuanqi along with related platforms. The account privileges and billing models vary depending on the entry point used.

Hy3 text and Agent model

Hy3 is the next-generation model released by Tencent in 2026, following the restructuring of its training and reinforcement learning infrastructure; it features a MoE architecture with 295 billion total parameters and 21 billion activation parameters, and supports 256K context length as well as multiple thinking modes. It is designed to enhance capabilities in complex reasoning, understanding of long texts, instruction follow-up, as well as code execution and tool invocation, making it suitable for use in Coding Agents, document automation, and multi-step workflows.

Hy3 has been integrated into Yuanbao, WorkBuddy, CodeBuddy, Marvis, and ima, and provides APIs through Tencent Cloud TokenHub. The model will continue to be updated; therefore, in production environments it is necessary to keep the Model ID fixed, as well as to save version information and evaluation records. It cannot be assumed that outputs with the same name will remain identical over time.

Tencent Yuanbao

Tencent Yuanbao is an AI assistant designed for individual users; it can handle tasks such as answering questions, conducting online searches, performing in-depth research, reading documents, writing text, and understanding multi-modal content. It can make use of the Hunyuan model or other models like DeepSeek, which means that the responses provided by Yuanbao are not necessarily all generated by the same version of Hunyuan.

The free offerings, model options, file size limits, and promotional benefits related to Yuanbao may change. Users should check the privacy settings before uploading contracts, work documents, and personal data.

Important conclusions must have their original sources verified.

Hunyuan’s multi-modal capabilities

1. Hunyuan Image Generation

HY-Image supports text-to-image generation, the use of multiple reference images, and image editing; it can be utilized for creating posters, product materials, conceptual designs, and content management. Tencent Cloud currently offers versions such as HY-Image V3.0 and Lite.

2. Hunyuan Video

The Hunyuan video model supports text-to-video and image-to-video generation; its open-source ecosystem includes HunyuanVideo along with various audio drivers, tools for person replacement, customization options, and research projects. Videos may still exhibit issues related to actions, text, physical laws, and consistency of the subjects, and permission must be obtained for the use of real people and brand materials.

3. Hunyuan 3D

HY-3D can generate grids, textures, and PBR materials from text or images, for use in games, e-commerce, design, AR/VR, and digital content. Hunyuan3D 2.x and later versions offer local inference capabilities, but the generated models still require verification regarding topology, overlap, number of faces, UVs, materials, and compatibility with the actual engine.

4. Visual and Embodied Intelligence

HY-Vision is used for image understanding, visual question answering, and multimodal reasoning; Tencent Hunyuan has also made public its research on world models, 3D scenarios, and embodied intelligence.

The performance of a model on a benchmark does not mean that it can safely control real devices; robot tasks require simulation, permission controls, and physical emergency stops.

Tencent Cloud API

Tencent Cloud Hunyuan API supports functions such as dialogue, content generation, long-text analysis, as well as handling of code, images, videos, and 3D data. The traditional Hunyuan interface relies on Tencent Cloud signatures and regional domain names, while TokenHub offers a unified entry point for accessing the models.

The authentication methods, Model ID, and billing may differ between the two approaches.

Keys should be stored only in the backend or a key management system. Agents and code calls need to have timeout, retry, concurrency, caching, content moderation, cost alerting, and manual confirmation mechanisms in place, to prevent unexpected costs arising from repeated tool calls.

Price and version comparison

Hy3 pay-as-you-go price

In the Guangzhou region, Tencent Cloud TokenHub shows that the price for online inference using Hy3 is:

• Input:1 yuan per million tokens;
• Output:4 yuan per million tokens;
• Cache hit:0.25 yuan per million tokens.

The new user page may offer 1 million tokens for free; the specific eligibility criteria and validity period are specified in the console. Prices may vary depending on the region, batch inference, enterprise contracts, and other Hunyuan models.

Hy Token Plan price

The Hy Token Plan is designed for individual developers and Agent workloads; the tokens generated by inputs, outputs, and caching are all deducted from the total amount allocated to the plan.

• Lite：28 yuan per month, 35 million tokens;
• Standard：78 yuan per month, 100 million tokens;
• Pro：238 yuan per month, 320 million tokens;
• Max：468 yuan per month, 650 million tokens.

The personal version currently allows the creation of only one API key, which is intended for individual development; teams should use the enterprise version.

The package only supports the models listed; attempting to use other models will result in an error.

The total number of tokens in a package cannot be simply converted into a fixed number of question-and-answer rounds; long contexts, toolkits, and multiple attempts all increase the consumption of tokens.

Image, 3D, and visual price examples

• HY-Image V3.0：0.2 yuan per sheet;
• HY-Image Lite：0.099 yuan per sheet;
• HY-3D 3.1：The score per task, based on its complexity, ranges from about 15 to 60 points.
• YT-VITA：Input: 1.2 yuan; output: 3.5 yuan per million tokens;
• HY-Vision 2.0 Instruct：The input cost is 7.5 yuan, while the output cost is 17.5 yuan per million tokens.

Failure in generation, security audits, resolution, duration, 3D quality levels, and asynchronous tasks may all have different charging rules; it is necessary to conduct a test with a small sample size before starting bulk production.

Tencent Yuanqi and enterprise applications

Tencent Yuanqi is the official platform for intelligent agents; it allows users to choose from models such as Hunyuan and DeepSeek, and to configure knowledge bases, workflows, and distribution channels. Companies can also use Tencent Cloud to create services for customer support, marketing, content moderation, data analysis, and industry-specific assistance.

Knowledge bases and agents cannot guarantee accuracy on their own. It is necessary to establish rules regarding permissions, versions, sources, and update strategies for the data, and to require manual approval for high-risk operations such as sending messages, calling external systems, writing to databases, and making payments.

GitHub and open-source models

Tencent has an official GitHub organization named Tencent-Hunyuan, whose open-source projects cover areas such as Hunyuan-A13B, Hunyuan-Large, HunyuanVideo, Hunyuan3D, image generation, OCR, world models, and embodied intelligence. Some of these repositories provide weights, inference code, tuning instructions, and technical reports.

Hunyuan’s open-source nature cannot be generalized; projects such as Hunyuan-A13B come with their own open-source or community licenses.

Hunyuan3D 2.x is governed by the Tencent Hunyuan community license; it has restrictions regarding the regions in which it can be used, requires additional application when the number of active users exceeds a certain threshold, imposes prohibitions on certain uses, mandates notification requirements for distribution, and includes limitations related to high-risk usage. It is not based on the standard MIT or Apache licenses.

The license description in the code file may differ from the latest LICENSE in the repository; the current complete agreement shall prevail.

Tencent Yuanbao, Hy3 cloud services, TokenHub, and the full commercial platform have not been made fully open source as a result. For local deployment, sufficient GPUs, video memory, as well as appropriate dependencies and security measures are required.

Tencent Hunyuan Usage Guide

Complete a basic task.

1. Clarify the purpose of the image, its dimensions, style, and requirements regarding commercial licensing.
2. Upload the original images for which you have usage rights to Tencent Hunyuan, or provide clear instructions.
3. Select the Hy3 text and the Agent model to generate low-cost candidates first;
4. Use Tencent Yuanbao to adjust the composition, the main subject, and the details;
5. Zoom in to examine faces, hands, text, edges, logos, and artifacts;
6. Export after confirming the color, resolution, and licensing, while retaining the original image;

Create reusable professional workflows

1. Create a list of brand colors, fonts, layout elements, and elements that are prohibited.
2. Test the multi-modal capabilities of the Hy3 text, the Agent model, Tencent Yuanbao, and Hunyuan separately;
3. Use the same set of representative samples to compare quality, speed, and cost;
4. Complex edges, text, and images of key products should be handed over for manual refinement.
5. Standardize naming, dimensions, and review status;
6. Batch processing and release are carried out after random inspections;

Data and copyright considerations

• Text, images, videos, and 3D outputs may all contain errors, deviations, or elements similar to those in the training data;
• Before publication, it is necessary to verify the facts, check for plagiarism, obtain permissions from relevant individuals, conduct trademark checks, and add AI content identifiers.
• The Open Model License does not automatically address the copyright of the output and the licensing of the input materials;
• Cloud-based calls handle prompts, files, and media;
• Companies should verify the location of the data, the retention period for logs, the sub-processors involved, as well as the mechanisms for training, use, and deletion of data. Sensitive information should be anonymized first, or enterprise solutions approved through contracts should be employed;

Which users is it suitable for?

• Individual users who use Tencent Yuanbao for Chinese-language question answering, searching, and reading documents;
• Developers who need Hy3 codes, long texts, and Agent capabilities;
• A content team that creates images, videos, and 3D assets;
• Operators who build knowledge base agents using Tencent Yuanqi;
• Organizations that require security, regional, and enterprise support from Tencent Cloud;
• Technical teams that research or deploy the Hunyuan open model locally.

Frequently Asked Questions

What is the difference between Tencent Hunyuan and Tencent Yuanbao?

Hunyuan is a family of models and technologies, while Yuanbao is an AI application designed for individuals. Yuanbao can make use of Hunyuan as well as provide other models.

Is Tencent Hunyuan free?

Yuanbao allows free use of some functions; Tencent Cloud may offer a free quota for new users.

API and Token Plans are paid services; the Hy Token Plan starts at 28 yuan per month.

How much is the Hy3 API?

Currently, the cost is 1 yuan for input, 4 yuan for output, and 0.25 yuan per million tokens for cache hits; prices vary by region and depending on any ongoing promotions, as indicated in the console.

Does Hunyuan support images, videos, and 3D?

Supported. Both Tencent Cloud and the open-source ecosystem offer models for image generation, video processing, 3D creation, and visual understanding; however, different access methods and models are charged separately.

Is Tencent Hunyuan open-source?

Some models have their code and weights made available publicly, but the complete products are not entirely open source. Each repository has its own licensing terms; for 3D models in particular, it is necessary to read the Tencent community licensing guidelines as well as the regional and commercial restrictions.

Can it be deployed privately?

The partially open model can be deployed locally in accordance with the relevant licenses, and Tencent Cloud also offers enterprise-level services. Self-deployment requires computational resources, content security measures, monitoring, and maintenance; it does not guarantee access to all the functions of Yuanbao or TokenHub.