Discover which prompt optimization techniques improve LLM performance by up to 200%. Benchmark of 5 methods, 3 models, and real datasets.