Tencent released the full version of its Hy4 large language model on November 4, 2026, 68 days after the August 28 release of Hy4 Preview, the date Caixin gives for Hunyuan's previous version. The company release says Hy4 adds visual understanding and computer use for the first time, keeps the Preview's API price, and will get open weights and system cards "in the coming days."
Tencent first announced Hy4 and Hy4 Flash on October 29, which we covered in Tencent Hy4 and Hy4 Flash Announced. This post covers the actual release and what Tencent published with it.
What Tencent says is new
According to Tencent's release, Hy4 "introduces visual understanding capabilities for the first time." Tencent lists fine-grained image processing, complex chart analysis and mathematical visual reasoning as tasks it now supports.
The second addition is computer use. Tencent says Hy4 can work with "both command-line interfaces and graphical user interfaces," and that it was trained on professional software such as Blender and CAD. Tencent also claims "major improvements in instruction following, long-context comprehension and long-horizon agentic tasks" over the Preview.
People's Daily Online quotes a Tencent Hunyuan official who says Hy4 "significantly expanded" (显著扩展了) the data and training scale in both pre-training and post-training compared with Hy4 Preview.
The size of the Preview is on record: Caiwen News reports that Hy4 Preview had 770 billion total parameters, 49 billion active parameters and a context length of more than 1 million tokens. Tencent's November 4 release does not give a parameter count for the final Hy4. People's Daily Online and Caixin both put the Hy4 series at 770B.
Tencent's own benchmark claims
All of the following figures come from Tencent and were published on November 4, 2026. None has been independently checked.
- Financial tasks. Tencent says a blind evaluation run by the Tencent Hy team with Tencent Financial Technology used 213 financial professionals and 579 real-world financial tasks. Tencent reports these scores: Hy4 3.04, DeepSeek-V4.1-Flash-1017 2.97, Kimi K3.1 Pro 2.68 and GLM-5.3 2.68. Tencent calls Hy4 "the top-performing open-source model" in this test.
- WorkBuddy. On WorkBuddy evaluation benchmarks, Tencent reports an average score of 83.18 and a pass rate of 96.15 percent, with Hy4 "ranking first place in nine out of twelve scenarios within the mainland Chinese market."
- Hallucinations. In internal evaluations, Tencent says Hy4's hallucination rate fell from 12.95% in Hy4 Preview to 7.2%, and errors caused by unsupported assumptions fell from 23.1% to 5.6%.
- Game development. Tencent says Hy4 improved by 19.1 percent on GameDev and 9.8 percent on GameLogic benchmarks compared with Hy4 Preview.
- Investment research. Tencent reports a 12.03 percentage-point improvement over Hy4 Preview on the internal iRaB benchmark run by financial AI company Rabyte.
Tencent's language is not all upbeat. Caixin reports that the Hunyuan team calls Hy4 "one of the strongest open-source models at present" (目前最强的开源模型之一), but also admits that Hy4 still has a large gap with top closed-source models. Caixin quotes the team: "Hunyuan will continue to steadily expand pre-training and post-training scale and strive to narrow the gap with top closed-source models."
Safety testing before the weights
Tencent ties the weights release to a safety review. According to the release, Hy4 is trained on defensive security data, and "offensive materials related to exploitation are explicitly excluded from the training dataset." Tencent says experts from its Xuanwu Lab and Keen Security Lab tested the model and "found no significant concerns regarding Hy4's offensive cybersecurity capabilities in the areas evaluated."
On biology, Tencent says internal expert reviews found that Hy4 "demonstrates strong foundational biological knowledge," but that "its capabilities in open-ended experimental design are more limited compared to frontier models."
Tencent said it plans to release the model weights and system cards "in the coming days." As of 4:16 PM PT on November 4, 2026, the Tencent account on Hugging Face listed only the Hy4-preview and Hy4-preview-FP8 repositories, both uploaded August 27, 2026 UTC.
Where Hy4 is available, and what it costs
Tencent says Hy4 is available now in Tencent Cloud TokenHub, WorkBuddy, CodeBuddy, Yuanbao, WeCom's AI Assistant (Dayuan) and ima. It says the API "is being made available via third-party platforms such as OpenRouter, Hermes, and OpenCode." OpenRouter already lists a tencent/hy4 model page; its model API shows the entry was created on November 3, 2026 PT. Tencent says integration with QQ, Honor of Kings, Game of Peace, WeRead and Tencent News is underway.
Tencent says Hy4 keeps the Preview's price: approximately US$0.90 per million input tokens, US$2.70 per million output tokens and US$0.045 per million cached input tokens. It also announced two promotions. WorkBuddy and CodeBuddy are free from November 4 to 15, and Hy4 is 50% off its standard price on Tencent Cloud TokenHub through December 31, 2026.
Hy4 Flash, the cheaper model Tencent named in October, has not shipped. The release says it "is scheduled for release soon." People's Daily Online reports that Hy4 Flash will "greatly reduce task costs" (大幅减低任务成本) while inheriting Hy4's capabilities.
Why it matters
Tencent is releasing on a fixed rhythm. Caixin reports that since Tencent rebuilt its AI infrastructure in February, Hunyuan has shipped a major version on average every two months, first as a Preview and then as an optimized full release. Caixin also reports that the Hy3 series was under 300B parameters and the Hy4 series is 770B, and that Tencent says Hy5 will be larger still.
For users, the practical changes are image input, computer use and the unchanged price. For the open-model field, the open question is the weights. Tencent's head-to-head numbers put Hy4 against DeepSeek, Kimi and GLM, which all have downloadable models. Until Tencent publishes Hy4's weights and system card, those comparisons rest on Tencent's own tests. For context on how these labs relate, see our guide to China's AI labs, and our explainer on Tencent Buddy One and WorkBuddy.
