Benchmarking Agentic LLM and VLM Reasoning for Gaming with NVIDIA NIM