Grok 4.5 Tops Tencent Hunyuan LHTB AI Agent Benchmark with 0.505 Average Score, 13 Tasks Completed

According to Tencent Hunyuan's latest test results, Grok 4.5 ranks first on the LHTB (Long Horizon Task Benchmark) with an average score of 0.505, completing 13 out of 46 tasks. The benchmark, developed jointly by Tencent Hunyuan and multiple universities, evaluates whether AI agents can continuously complete complex tasks on local endpoints.

LHTB covers 46 tasks spanning software engineering, experiment reproduction, scientific computing, and multimodal analysis, with each task running up to 90 minutes and typically requiring hundreds of consecutive steps. Of the 20 models publicly tested, 29 tasks remain uncompleted by any model, indicating the benchmark's high difficulty.