The rapid development of large language models (LLMs) has brought new opportunities for enhancing multi-agent collaborative systems. However, most existing LLM-based multi-agent frameworks often lack depth in cross-domain knowledge integration and multi-step reasoning, or exhibit insufficient comprehensiveness for complex problems. Inspired by multi-team collaboration models, we propose NexaFusion, a multi-team collaboration framework optimized for complex tasks. NexaFusion enables multiple teams to independently generate diverse solutions through task decomposition, role optimization, and dynamic integration mechanisms, with iterative optimization and integration ensuring accurate and innovative final results. We comprehensively evaluated NexaFusion on three complex tasks: creative writing, open-ended question answering, and logical reasoning, also exploring the impact of team size on experimental results. Findings demonstrate NexaFusion outperforms existing methods in generating high-quality solutions and excels in handling solution diversity and complexity. Through flexible collaboration and efficient integration, NexaFusion provides innovative solutions for cross-domain knowledge integration and reasoning in multi-agent collaborative frameworks.