OpenEvolve: Teaching LLMs to Discover Algorithms Through Evolution

OpenEvolve combines large language models with evolutionary algorithms to automatically discover and optimize code, using a quality-diversity framework where AI models generate program variations that are evaluated and refined across multiple isolated populations. The system has achieved breakthrough results including 321x speedups through automatic discovery of JAX compilation, GPU kernel optimizations for Apple Silicon, and 10%+ improvements in prompt optimization across various benchmarks.