Tencent Launches Hy4 with Image Understanding and Computer Use
Tencent has launched Hy4, the latest version of its Hunyuan AI model, introducing image understanding and computer-use capabilities alongside improvements in coding, game development, financial analysis, and long-running tasks.
👋Hy4 is here, new with vision.
Firmly among the leading open models.
More productive everywhere: finance, frontend, game, officework, research…
Built responsibly, with cyber/biosecurity in mind.
Try it out and let us know!
🔗：https://t.co/ZzL4DtgH3h pic.twitter.com/xceRxN6ueT– Tencent Hy (@TencentHunyuan) November 4, 2026
Announced on November 4, 2026, Hy4 builds on the Hy4 Preview released in August. The new version goes beyond text-based interactions, allowing it to analyze visual content and interact with computer environments through command-line and graphical interfaces. Tencent has also announced plans to release the model weights and system card in the coming days.
The update positions Hy4 as a more versatile model for developers, researchers, and businesses looking to automate work across different applications.
Image Understanding and Visual Reasoning
One of the main additions in Hy4 is its ability to understand and analyze images, documents, diagrams, and other visual content.
The model can extract information from images, examine complex charts, and perform mathematical reasoning based on visual inputs. This goes beyond basic image descriptions or object recognition. Tencent says its training focuses on interpreting fine details and connecting visual information with broader reasoning tasks.
These capabilities could be useful when working with reports containing charts and statistics, technical diagrams, or documents that combine text and images. For researchers, students, and analysts, this means Hy4 can help interpret information presented visually rather than relying only on written prompts.
Computer Use: Interacting with Applications and Tools
Another major addition is Computer Use, which allows Hy4 to interact with computers through two interfaces: the command line (CLI) and graphical user interfaces (GUI).
Instead of only generating code or explaining how to complete a task, the model can work with tools and applications inside a computer environment, carrying out sequences of actions as part of a larger workflow.
Tencent says it has provided specialized training for professional software, including Blender and computer-aided design (CAD) tools used in engineering and design. This training focuses on tasks that require planning and multiple steps rather than a single command or response.
For example, these capabilities can support workflows involving 3D scenes in Blender, engineering designs, and tasks that require switching between different applications. The broader goal is to connect the model’s understanding of a task with its ability to interact with the software needed to complete it.
Coding and Game Development Improvements
Tencent has also focused on complex software development tasks. According to the company, Hy4 received additional training for projects involving business logic, state management, and coordination between front-end and back-end components.
Game development is another area covered by the update. The Hunyuan team worked with developers from Tencent Interactive Entertainment Group, training the model on game engines such as Unreal Engine, Unity, and Godot.
The work covers game logic and multiplayer systems, where different components must function together rather than operate as isolated pieces of code.
Tencent reports that Hy4 improved its score by 19.1% on the GameDev benchmark and 9.8% on GameLogic compared with Hy4 Preview. These figures come from the company’s evaluations and indicate progress on game development and game logic tasks.
Hy4 Benchmarks: Financial Analysis and Fewer Errors
Tencent published several evaluation results to demonstrate the improvements in Hy4. One of the highlighted tests focused on financial analysis and was conducted in collaboration with Tencent Financial Technology.
The evaluation included 579 real-world financial tasks and 213 specialists, including 153 experts from external institutions. Hy4 achieved a score of 3.04, ahead of the competing models included in the same assessment.
| Model | Evaluation Score |
|---|---|
| Tencent Hy4 | 3.04 |
| DeepSeek V4.1 Flash-1017 | 2.97 |
| Kimi K3.1 Pro | 2.68 |
| GLM-5.3 | 2.68 |
The result reflects performance on the financial tasks included in this particular evaluation. It should not be interpreted as a universal ranking of AI models across all benchmarks.
Tencent also reported a 12.03 percentage-point improvement over Hy4 Preview on the Investment Research Agent Benchmark, conducted by financial AI company Rabyte.
Another focus of the update is reducing inaccurate responses and unsupported assumptions. According to Tencent’s internal evaluations, the hallucination rate fell from 12.95% in Hy4 Preview to 7.2% in Hy4. Errors associated with unsupported assumptions also declined from 23.1% to 5.6%.
WorkBuddy, Office Productivity, and Scientific Research
Hy4 also powers improvements in Tencent’s WorkBuddy assistant, which supports office workflows such as document and presentation creation, creative tasks, 3D scene generation, and web development.
In WorkBuddy evaluations, Hy4 achieved an average score of 83.18 and a 96.15% pass rate. Tencent says the model ranked first in nine out of 12 scenarios in the assessment conducted in the Chinese market.
The company has also added integration with WeCom, connecting the assistant to everyday workplace information, including conversations, email, documents, calendars, and meetings. This integration can help users track projects, identify unfinished tasks, and monitor upcoming meetings and deadlines.
In finance, Hy4 received specialized training for fund research, securities analysis, and quantitative investment strategies. Tencent says this training draws on thousands of specialized workspaces and tens of thousands of tasks based on real-world scenarios.
Scientific research is another area covered by the model’s reasoning and coding capabilities. Tencent cited the TenureTrail evaluation, which included 123 executable tasks across seven fields, including mathematics, physics, environmental science, systems engineering, and social economics.
Hy4 Pricing and Availability
Hy4 is available through several Tencent services, including Tencent Cloud TokenHub, WorkBuddy, CodeBuddy, Yuanbao, WeCom AI Assistant, and ima. The model has also become available through external platforms such as OpenRouter, with additional integrations in development.
Tencent says Hy4 retains the pricing structure of Hy4 Preview, with approximate rates of $0.90 per million input tokens, $2.70 per million output tokens, and $0.045 per million cached tokens.
OpenRouter lists more precise base prices of $0.834 per million input tokens, $2.501 per million output tokens, and $0.042 per million cached tokens. Its displayed promotional rates are lower, at $0.417 for input tokens and $1.251 for output tokens, reflecting a 50% discount.
Tencent Cloud TokenHub’s 50% promotional offer runs through December 31, 2026. Tencent has also announced free access to WorkBuddy and CodeBuddy from November 4 to November 15, subject to each service’s usage limits and conditions.
Are Hy4 Model Weights Available for Download?
Tencent says it plans to release Hy4’s model weights and system card in the coming days. The release will give developers and researchers more information about the model and open the door to using it outside Tencent’s hosted services.
For now, access to the new version is available through cloud services and APIs. Tencent has not yet announced the direct download of Hy4’s model files.
This is different from Hy4 Preview, whose weights were previously released for developers to run in their own environments, subject to hardware requirements and the applicable license. For the new version, local deployment and commercial-use conditions will depend on the weights and official licensing terms Tencent publishes.
Security and Safety Measures
Tencent has also addressed security in its announcement, emphasizing defensive cybersecurity applications. The company says its training data covered vulnerability analysis and software remediation, while excluding material related to building hacking tools or breaking into systems.
Experts from Tencent’s Xuanwu Lab and Keen Security Lab conducted security evaluations covering reverse engineering, executable-file analysis, and vulnerability testing. Tencent reported that these assessments did not identify major concerns about the model’s offensive capabilities within the areas examined.
On biological safety, the company said Hy4 has a solid foundation in general biological knowledge but remains less capable than leading models when handling open-ended experimental design.
With Hy4, Tencent is expanding its Hunyuan model beyond the capabilities introduced in the Preview release. Image understanding and computer interaction are the most significant additions, while improvements in coding, finance, and office productivity broaden its potential applications.
The upcoming release of the model weights and system card will be an important next step, particularly for developers and researchers who want to examine the model and run it outside Tencent’s services.
