Reaching SOTA Without Breaking the Bank: Using AI21 Maestro to optimize deep research agents

Developing AI agents for production involves a difficult trade-off between accuracy, cost, and latency, where improving one often compromises the others. AI21 Maestro is an optimization framework that systematically navigates this complex challenge by automatically exploring countless configurations. The framework identifies the most efficient combinations of models, tools, and scaling strategies, presenting a full frontier of optimal performance points. This allows teams to select their ideal balance between budget and quality, a process that achieved state-of-the-art accuracy on the BrowseComp-Plus research benchmark.