Large Language Models (LLMs) have achieved remarkable success in Natural Language Processing (NLP) tasks, including Knowledge Graph Construction (KGC). Recent studies have explored using LLMs to automate KGC, yielding promising results. Existing approaches typically use a single LLM, and design task-specific prompts to instruct it. However, this often results in high computational demands and overlooks the varying capabilities of different models across tasks. Moreover, applying LLM solutions in practical scenarios requires a more complex utility structure beyond performance and cost. Therefore, this doctoral thesis investigates the strategic planning of multiple LLMs, taking the KGC as a use case, aiming to enhance overall performance while ensuring cost efficiency.