Tag
LoopArena introduces a benchmark to evaluate models acting as runtime controllers for loop engineering in coding agent tasks, revealing low strict success rates but significant cost reductions.