←Story detail

Trending story Continuously updated

Arena.ai teased the upcoming release of Tencent Hunyuan Hy4, and Tencent Hunyuan subsequently launched the official version

12 reports6 discussions18 story updatesUpdated Nov 5 00:05

Understand this story

AI overview

On November 4, 2026, Arena.ai teased the upcoming release of Tencent Hunyuan's next-generation model Hy4 and cited the preview version's evaluation performance on Agent Arena: in evaluations of more than 14,500 real agent task sessions, the Hy4 preview ranked second among open models and tenth overall worldwide, with a net improvement rate of 5.4% and a median cost per individual task of about $0.26, roughly 68% lower than Kimi K3 (Max)'s $0.80. On the same day, Tencent Hunyuan released the official version of Hy4, saying that compared with the preview version it significantly expanded the scale of pretraining and post-training, introduced visual capabilities for the first time and supported computer operation for the first time, improving by 19.1% and 9.8% on the GameDev and GameLogic evaluations respectively, while the hallucination rate dropped from 15.76% to 5.45%. Hy4 is now available on Tencent Cloud and OpenRouter, can be experienced in products such as WorkBuddy/CodeBuddy, Yuanbao and ima, with API pricing remaining unchanged, and there are plans to open-source the model weights.

AI-generated from public reports

Latest development · Report

Tencent Releases Hy4 Frontier Large Model, Adding Visual Understanding and Computer Operation Capabilities for the First Time

Developments

2 confirmed developments

  1. Arena.ai teases the upcoming release of Tencent Hunyuan Hy4

    Source:Phoenix Technology
  2. Tencent Hunyuan releases official version of Hy4

    Source:English text is being prepared.

Reports and discussion timeline

Browse reports and discussion by time. Their evidence levels remain distinct.

Nov 5

  1. SiliconANGLESelected

    Tencent Releases Hy4 Frontier Large Model, Adding Visual Understanding and Computer Operation Capabilities for the First Time

    Tencent has released the large language model Hy4, which for the first time adds computer operation capabilities, combining visual understanding, reasoning, and tool calling to achieve long-horizon agent performance. The model, like the August 28 Hy4 Preview, has 770 billion parameters, 49 billion activated parameters, and a 1 million token context window, and improves by 19.1% and 9.8% on the GameDev bench and GameLogic benchmarks respectively, though the baseline values were not disclosed. The model is now available on platforms including Tencent Cloud TokenHub, CodeBuddy, and Yuanbao, and can also be accessed through third-party channels such as OpenRouter, with pricing maintained at approximately 90 cents per million input tokens, $2.70 for output, and 4.5 cents for cached input; the Flash version and model weights will be released in the coming days.

Nov 4

  1. YicaiRelated discussion

    Yicai joined this discussion

    This post contributes to heat; it is not a formal report or confirmed fact.

  2. Shao MengRelated discussion

    Shao Meng joined this discussion

    This post contributes to heat; it is not a formal report or confirmed fact.

  3. Adina YakupRelated discussion

    Adina Yakup joined this discussion

    This post contributes to heat; it is not a formal report or confirmed fact.

  4. Shao MengSelected

    Tencent Hunyuan Releases Hy4: First to Support Vision and Computer Use, Weights Planned to Be Open-Sourced

    Tencent Hunyuan has officially released Hy4, with approximately 770B total parameters, a MoE architecture, and a 1M token context. It is the first to add visual understanding and Computer Use capabilities, and can operate command lines and graphical interfaces. In financial evaluations, Hy4 scored 3.04 on a four-point scale, slightly above DeepSeek-V4.1-Flash-1017's 2.97, and its search-related hallucination rate dropped from 15.76% in the preview version to 5.45%. The API is currently available through entry points such as TokenHub, OpenRouter, and Yuanbao, at 6 yuan per million input tokens and 18 yuan per million output tokens. The weights are planned to be open-sourced but have not yet been released.

  5. Tencent WorkBuddyRelated discussion

    Tencent WorkBuddy joined this discussion

    This post contributes to heat; it is not a formal report or confirmed fact.

  6. Tencent HunyuanFirst-partySelected

    Tencent Hunyuan Releases Hy4 Model, Adding Vision Capabilities

    Tencent Hunyuan releases the Hy4 model, adding vision capabilities, and the official statement says it stands firmly among the leading open-source models. The model is aimed at boosting productivity in scenarios such as finance, frontend, gaming, office work, and research, and was designed with cyber and biosecurity in mind. The official team provided a trial link.

  7. Testing Catalog

    Tencent Releases Flagship Model Hy4, Now Available on Tencent Cloud, OpenRouter and Other Platforms

    Tencent has released its new flagship model Hy4, claiming its benchmark performance ranks among the top open-source models. Hy4 is now available on Tencent Cloud, OpenRouter, CodeBuddy, WorkBuddy and Hy AI Studio, with open weights coming soon. Tencent plans to open-source the Hy4 model weights and has partnered with the vLLM (Inferact Inc.) and SGLang communities for ecosystem support; when the weights are released, a system card will also be published detailing technical details such as safety evaluations.

  8. IT Home · Artificial Intelligence

    Tencent Hunyuan Hy4 Official Version Released, Introducing Vision and Computer Operation Capabilities for the First Time

    Tencent Hunyuan has released the official version of the Hy4 large language model. Compared with Hy4 preview, it significantly expands the scale of pre-training and post-training, introduces vision capabilities for the first time, and supports computer operation, enabling it to complete tasks by combining command-line and graphical interfaces. Hy4 is now available on Tencent Cloud and OpenRouter, and can be experienced in products such as CodeBuddy / WorkBuddy, Yuanbao, and ima, while maintaining the same API pricing as the preview. The official team plans to open-source the model weights and release a model system card in the near future, while also stating that Hy4 still has a considerable gap compared with top closed-source models on highly difficult test sets such as frontier mathematics, long-horizon development, and visual question answering.

  9. IT Home · AIRelated discussion

    IT Home · AI joined this discussion

    This post contributes to heat; it is not a formal report or confirmed fact.

  10. ArenaRelated discussion

    Arena joined this discussion

    This post contributes to heat; it is not a formal report or confirmed fact.

  11. Tencent HunyuanFirst-party

    Tencent Hunyuan Releases Hy4, Adding Visual Capabilities

    Tencent Hunyuan announced the launch of Hy4, adding visual capabilities, and stated that it ranks firmly among leading open-source models, is more efficient in scenarios such as finance, front-end, gaming, office work, and research, and was designed with cyber and biosecurity in mind.

  12. ZhidongxiSelected

    Tencent Releases Hy4 Official Version: First Introduction of Visual Capabilities, Hands-On Testing of Operating WeChat and 3D Modeling

    On November 4, Tencent released the latest flagship model Hy4 official version and plans to open-source the model weights. Compared with the Hy4 preview in August, it significantly expands the scale of pre-training and post-training and introduces visual capabilities for the first time. Hy4 is now available on Tencent Cloud and OpenRouter and can be experienced in products such as CodeBuddy/WorkBuddy, Yuanbao, and ima, with API pricing comparable to the DeepSeek V4 Pro range. Zhidongxi's advance internal testing found that, with native visual understanding now in place, it can directly identify modification points from screenshots; in 3D modeling tasks, it can still maintain the original goal even after using up a 500,000-token context window; and operating the WeChat desktop client to post comments and summarizing group chat history by scrolling up can also run successfully.

  13. Phoenix TechnologySelected

    Tencent Releases Official Version of Flagship Model Hy4: Introduces Vision Capabilities for the First Time and Plans to Open-Source Weights

    Tencent has released the official version of its new-generation flagship model Hy4. Compared with the Hy4 preview in August, it has significantly expanded the scale of pre-training and post-training, introduces vision capabilities for the first time, and plans to open-source the model weights. Hy4 is now available on Tencent Cloud and OpenRouter and can be experienced in products such as CodeBuddy/WorkBuddy, Yuanbao, and ima, with API pricing unchanged from Hy4 preview. Zhidongxi's hands-on testing shows that Hy4 can modify webpages from images, model in Blender, and operate the WeChat desktop version to post comments and summarize group chats. Its goal-retention capability in long-horizon tasks is relatively strong, but execution is somewhat slow and it takes over the mouse.

  14. Tencent Hunyuan ResearchFirst-partySelected

    Tencent Hunyuan Releases Hy4 Official Version, Introducing Vision and Computer Operation Capabilities for the First Time

    Tencent Hunyuan Releases Hy4 Official Version, Significantly Expanding Pre-training and Post-training Scale Compared with Hy4 preview, Introducing Vision Capabilities and Supporting Computer Operation for the First Time. The company says Hy4 has for the first time ranked among the strongest open-source models, improving by 19.1% and 9.8% on the GameDev and GameLogic benchmarks respectively, while the hallucination rate dropped from 15.76% to 5.45%. Hy4 is now available on Tencent Cloud and OpenRouter and can be experienced in products such as WorkBuddy/CodeBuddy, Yuanbao, and ima. API pricing remains unchanged at 6 yuan for input and 18 yuan for output per million Tokens, and the model weights are planned to be open-sourced.

  15. Caixin

    Tencent Releases Hunyuan Hy4 Large Model with 770B Parameters and First Introduces Visual Understanding

    On November 4, Tencent released its new-generation large language model Hunyuan Hy4, with practicality as its core principle, and will open-source the model weights in the near future. The Hy4 series has a parameter scale of 770B, compared with less than 300B for the previous-generation Hy3 series, and Tencent says the next version, Hy5, will be even larger. The Hunyuan team claims that Hy4 is one of the strongest open-source models currently available, while acknowledging that it still has a considerable gap with top closed-source models.

  16. Phoenix Technology

    Tencent Hunyuan Hy4 to Be Released Soon: Preview Version Ranks 10th Globally on Agent Arena, Task Cost Lower Than Kimi…

    Arena.ai revealed that Tencent Hunyuan's next-generation model Hy4 is about to be released. Its preview version ranked second among open models and 10th overall globally in Agent Arena's evaluation of more than 14,500 real agent task sessions, second only to Kimi K3 (Max). The Hy4 preview version had a net improvement rate of 5.4%, compared with 6.6% for Kimi K3 (Max); its median cost per task was about $0.26, roughly 68% lower than Kimi K3 (Max)'s $0.80, and it entered the Pareto frontier of that leaderboard. Tencent Hunyuan has not yet announced the release date, model parameters, or full technical details of the official version.

  17. Arena

    Tencent Hunyuan Hy4 is about to land on the Arena leaderboard

    Arena.ai has teased that Tencent Hunyuan's Hy4 is about to be released, and said it will be watching its ranking on the Arena leaderboard. The quoted post recalled that last month the Hy4 preview ranked 2nd in the Agent Arena open-source lab, ranked 2nd among open-source models (10th overall), with a net improvement of +5.4%, second only to Kimi K3 (Max)'s +6.6%; the cost per median task was $0.26, about 68% lower than Kimi K3 (Max)'s $0.80.

Heat trend for this story

Current
90
Period peak
90 Nov 5 03:00
24-hour change
—

Move the pointer or select the chart to inspect hourly heat. Use the left and right arrow keys with a keyboard.

Related stories

  1. Related storiesMistral releases Large 4 weights on Hugging Face
  2. Related storiesOpenAI launches GPT-6.1 Terra on ChatGPT Work and Codex
  3. Related storiesMiniMax releases the M3.1 Flash model
  4. Related storiesMeta launches Muse Wonder and Muse Wonder Full models