optimize_anything: A Universal API for Optimizing any Text Parameter

Researchers introduced optimize_anything, an open-source LLM-based optimization framework that turns many problems into scoring-based text editing tasks.

It reported state-of-the-art or better-than-specialized results on six benchmarks, including agent discovery, cloud scheduling, CUDA kernel generation, and circle packing.

The system also showed that richer feedback and multi-task search can improve performance and transfer across tasks.

The work suggests a single text-based optimization approach may work across very different domains, from AI agents to infrastructure and code generation.