GPT-6 showed little performance improvement on existing financial benchmarks due to saturated evaluation tasks, ambiguous prompts, and erroneous answer keys