- The Hy4 model launched by Tencent is capable of computer operations and visual understanding.
- Hy4 has performed exceptionally well in multiple testing scenarios, integrating with WeCom to enhance project management.
- The model received high scores in the field of financial research, demonstrating its professional capabilities.
- Tencent is committed to reducing the factual error rate of Hy4 and emphasizes cybersecurity measures.
Tencent has launched a new open-source AI model called Hy4, which is capable of operating computers, interpreting complex images, and handling challenging engineering tasks. This model marks Tencent's latest attempt to compete with leading open-source AI systems. Hy4 adds visual understanding capabilities to Tencent's Hy series of models for the first time, enabling it to process detailed images, interpret complex charts, and solve mathematical problems involving visual information. The model also introduces computer usage capabilities, allowing it to interact with command-line tools and graphical interfaces.
Tencent plans to release the model weights and system card for Hy4 in the coming days and has stated that it has incorporated security measures against cyber threats and biological risks.
Hy4's Performance in Testing
Hy4's proficiency in software operation is impressive, including applications like Blender and computer-aided design (CAD). Tencent trained the model to handle complex workflows that require planning and multiple execution steps, which can help automate tasks that traditionally require significant manual input. The model also supports 3D scene creation, website development, and document generation. Tencent reported an average score of 83.18 on its WorkBuddy platform, with a pass rate of 96.15% for Hy4 in assessment tasks.
In 12 testing scenarios in mainland China, Hy4 ranked first in 9 of them. WorkBuddy integrates Hy4 with Tencent's collaboration software WeCom, allowing the model to access information from chats, emails, documents, calendars, and meetings to track project progress, identify incomplete tasks, and mark upcoming deadlines.
Applications in Financial Research
Financial research is another major application area for Hy4. Tencent trained the model using thousands of professional workspaces and tens of thousands of professional tasks, covering fund research, securities analysis, and quantitative investment strategies. In a blind assessment involving 213 finance professionals, Hy4 received the highest score. This assessment included 579 real-world financial tasks, and Tencent reported that about 70% of participants had over five years of industry experience. Hy4 scored 3.04, slightly ahead of DeepSeek-V4.1-Flash-1017's 2.97, while Kimi K3.1 Pro and GLM-5.3 both scored 2.68.
Software development was also a focus during training, with Hy4's GameDev benchmark score improving by 19.1% compared to Hy4 Preview, and the GameLogic score increasing by 9.8%. The model supports development workflows involving Unreal Engine, Unity, and Godot, with Tencent stating that these improvements extend to complex game mechanics and multiplayer implementations.
Tencent is also committed to reducing Hy4's factual error rate when retrieving information. Internal evaluations showed its hallucination rate decreased from 12.95% to 7.2%, and errors related to unsupported assumptions dropped from 23.1% to 5.6%. Cybersecurity measures are another aspect of this release, with Tencent training Hy4 on defensive security materials, including vulnerability analysis and patching. According to the company, this training does not include offensive exploitation materials. Hy4 is available through Tencent Cloud TokenHub, WorkBuddy, CodeBuddy, Yuanbao, and several other platforms, with its API also supporting third-party services, including OpenRouter and OpenCode.
Despite the reported performance improvements, Hy4's pricing remains the same as Hy4 Preview, with Tencent charging approximately $0.90 per million input tokens, about $2.70 per output token, and around $0.045 for cached input million tokens.
Tencent also plans to soon launch Hy4 Flash, a smaller product aimed at applications where operating costs are particularly important.
Diverse Applications and Market Potential of the Hy4 Model
Tencent's Hy4 model not only demonstrates exceptional capabilities in computer operations but has also received positive evaluations in areas such as visual understanding and financial research. This indicates its potential for automation and handling professional tasks, which can enhance work efficiency and reduce labor costs. Additionally, Tencent's emphasis on cybersecurity adds credibility to its products, especially in the context of the accelerating digital transformation, where businesses increasingly demand security. The diverse applications of Hy4 and its competitiveness in the market may lead the future development direction of AI technology.

