Tencent Releases Hy4, Adding Vision and Computer Use and Saying It Now Ranks Among the Strongest Open Models
👋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! 🔗：
Context
A large Chinese lab says its new model is among the strongest open models and plans to publish the weights, which would give builders another capable model with vision and computer use they can run themselves.
Hy is Tencent's family of AI models, and its X account is @TencentHunyuan. Tencent released an earlier Hy4 preview, which it uses as the baseline for most comparisons. Other leading open weight models from China include DeepSeek, Kimi and GLM, which Tencent compares against in its chart.
Deep dive
Tencent's Hy team (@TencentHunyuan) announced Hy4 early on Nov 4 ET, saying it is firmly among the leading open models and adds vision for the first time.
According to Tencent's announcement page, Hy4 substantially scales up pre training and post training compared with Hy4 preview. Tencent says it adds image understanding and, for the first time, computer use across command line and graphical interfaces, with targeted training on software such as Blender and CAD. Tencent also reports gains in instruction following, long context understanding, long horizon tasks and deep research.
Tencent cites several internal results. It says Hy4 improves by 19.1% on GameDev over Hy4 preview, cut a search hallucination rate from 15.76% to 5.45%, and ranked first among open weight models on 579 finance tasks judged with 213 finance professionals, scoring 3.04 against 2.97 for DeepSeek V4.1 Flash 1017. These figures come from Tencent and its partners and have not been independently verified.
The chart in the post compares Hy4 with GLM 5.3, DeepSeek V4.1 Flash, Kimi K3.1 Pro, Claude Opus 5.5 and GPT 6.1 Sol. Hy4 beats GLM 5.3 and DeepSeek V4.1 Flash on every chart where they appear, but Claude Opus 5.5 scores higher than Hy4 on all twelve benchmarks shown. Tencent itself says a notable gap remains with top proprietary models.
The weights are not out yet. Tencent says it plans to open source them, working with the vLLM and SGLang communities, and will publish a system card then. For now Hy4 is available on Tencent Cloud, OpenRouter and Tencent products such as CodeBuddy, WorkBuddy and Yuanbao, priced at $0.834 per million input tokens and $2.501 per million output tokens.
What to expect next
The key thing to watch is when Tencent actually releases the Hy4 weights and system card, since the open model claim depends on it. Independent benchmark groups will likely test Hy4 against DeepSeek, Kimi and GLM. Tencent also says a cheaper and faster Hy4 flash is coming soon. Builders will test whether the computer use features hold up outside Tencent's own demos.