Woofun AI reports that Grok 4.5 has secured the top position on the Tencent MetaUnity LHTB leaderboard, achieving an average score of 0.505 by successfully completing 13 out of 46 designated tasks. The benchmark evaluates AI agents' ability to perform complex operations in terminal environments, with each task capped at 90 minutes and requiring hundreds of sequential steps across domains like software engineering and scientific computing.

The assessment methodology employs a hidden validator to verify file outputs and test results, disregarding self-reported completion status. While GPT-5.5 previously led early tests with seven completed tasks, the expanded public dataset of 20 models shows Grok 4.5 as the clear leader.

However, significant challenges remain, as 29 tasks have not been completed by any model, with most agents failing due to time exhaustion or premature termination.