Tenfold Capacity Expansion After Surge in Tencent Hy4 Invocations; No. 1 Among Open-Source Models in Agent Capabilities

On November 5, Tencent announced that WorkBuddy had expanded tenfold in response to a surge in Hy4 invocations, in order to meet the experience and usage needs of more users. Meanwhile, driven by continued improvements in model capabilities, Hy4 received strong recognition on internationally authoritative model evaluation leaderboards, ranking second among Chinese models on the Artificial Analysis Intelligence Index and first among open-source models on the Agent Arena leaderboard.

Since Hunyuan Hy4 was first made available through WorkBuddy, its significantly enhanced Agent capabilities have been actively experienced and enthusiastically received by a broad range of users and developers, with many users reaching their usage quotas on the first day of launch. Tencent has now expanded the Hy4 inference cluster tenfold, increased users’ free trial quotas, and continued to dynamically allocate resources.

In line with the enthusiastic response from users, Hy4 delivered impressive results across several authoritative overseas evaluations. On the 5th, the globally authoritative large-model evaluation platform Artificial Analysis released its latest Intelligence Index leaderboard, in which Tencent Hy4 ranked among the leading models out of 724 and placed second among Chinese models, behind only Kimi K3.1 Pro and ahead of models such as Qwen 3.8 Max and DeepSeek V4.1 Flash.

Artificial Analysis is regarded within the industry as an important reference for measuring the comprehensive intelligence of large models. The index is not a single benchmark score, but a high-difficulty evaluation that combines core capabilities such as complex reasoning, code generation and debugging, tool use, multimodal understanding, long-horizon planning, and scientific problem solving, placing substantial demands on models’ generalization capabilities and engineering practicality. Hy4 tied with Grok 4.7 (xhigh) in this comprehensive evaluation system, demonstrating Tencent’s solid foundation in foundation-model capabilities.

Artificial Analysis also noted that, in terms of cost efficiency, Hy4 costs approximately US$0.78 per task, significantly less than Grok 4.7 (xhigh) and Qwen3.8 Max within the same score range, with prices 10 times and 13 times cheaper than the two models, respectively, demonstrating a strong balance between capability and cost.

Furthermore, on the latest Agent Arena leaderboard, which places greater emphasis on performance in real-world agent tasks, Tencent Hy4 ranked 13th overall and first among open-source models worldwide. With its high cost-performance ratio, Hy4 also sits on the leaderboard’s Pareto Frontier, maintaining a positive score while significantly reducing costs compared with top models such as the more expensive Claude Opus 5.5.

On November 4, Tencent Hunyuan released Hy4, its latest-generation large language model, delivering another comprehensive leap in model capabilities. Compared with Hy4 preview, Hy4 significantly expanded the scale of data and training during both pretraining and post-training, introduced visual understanding capabilities for the first time, and greatly improved overall performance in instruction following, long-text comprehension, and long-horizon agents.

This article was provided by Tencent. QbitAI has been authorized to reprint it, and all views belong to the original author.