Text this: Benchmarking large language models GPT-4o, llama 3.1, and qwen 2.5 for cancer genetic variant classification