“The top five to seven large language models have converged in capability and are now largely interchangeable with minimal differentiation.”