Skip to content
Main Site News Console

Tencent Hy4 Expands Capacity 10x Amid Surging Usage, Ranks #1 in Open-Source Agent Capabilities

· 量子位
国内AI

Tencent Announces 10x Capacity Expansion for WorkBuddy Amid Surge in Hy4 API Calls

On November 5, Tencent announced that WorkBuddy has completed a tenfold capacity expansion in response to a surge in Hy4 usage, aiming to meet growing demand from users and developers. At the same time, driven by continuous leaps in model performance, Hy4 has earned high acclaim on authoritative global LLM benchmarks, ranking second among Chinese models on the Artificial Analysis Intelligence Index and first among open-source models on Agent Arena.

Since its debut integration into WorkBuddy, Hunyuan Hy4 has garnered enthusiastic feedback and high engagement from users and developers thanks to its significantly improved agent capabilities. On launch day, a large number of users hit their quota limits. In response, Tencent completed a 10x capacity expansion of its Hy4 inference cluster, raised the free trial quota for users, and plans to dynamically allocate compute resources going forward.

Echoing the enthusiastic reception from users, Hy4 has also delivered impressive results across several authoritative international benchmarks. On November 5, global LLM evaluation platform Artificial Analysis released its latest Intelligence Index leaderboard. Tencent’s Hy4 ranked near the top among 724 evaluated models, placing second among Chinese models—behind only Kimi K3.1 Pro and ahead of models like Qwen 3.8 Max and DeepSeek V4.1 Flash.

Image 3

Artificial Analysis is widely regarded in the industry as a key benchmark for evaluating the overall intelligence of large language models. Rather than relying on a single metric, the index is a rigorous, comprehensive evaluation spanning core capabilities such as complex reasoning, code generation and debugging, tool use, multimodal comprehension, long-horizon planning, and scientific problem-solving. This places extremely high demands on a model’s generalization ability and real-world engineering utility. Hy4 tied with Grok 4.7 (xhigh) within this evaluation framework, underscoring Tencent’s solid foundation in foundational model research and development.

Artificial Analysis also noted that in terms of cost efficiency, Hy4’s per-task cost is approximately $0.78—significantly lower than Grok 4.7 (xhigh) and Qwen 3.8 Max within the same capability tier. Hy4 is 10x and 13x cheaper than these two models respectively, demonstrating an exceptional balance between performance and cost.

Furthermore, on the latest Agent Arena leaderboard—which focuses on evaluating performance across real-world agent tasks—Tencent Hy4 ranked 13th overall and 1st among open-source models globally. Thanks to its remarkable cost-performance ratio, Hy4 also made the benchmark’s Pareto Frontier, maintaining competitive positive scores while dramatically cutting usage costs compared to expensive top-tier models like Claude Opus 5.5.

Image 4

Image 5

On November 4, Tencent Hunyuan officially launched its next-generation LLM, Hy4, marking an across-the-board leap in model capabilities. Compared to the Hy4 Preview version, Hy4 significantly expanded its data and training scale across both pre-training and post-training stages, introduced visual comprehension capabilities for the first time, and substantially boosted its overall performance in instruction following, long-context understanding, and long-horizon agent scenarios.