Agent swarms and the new model economics

Researchers developed an improved agent swarm system that decomposes complex tasks into hierarchical trees, with planner agents delegating work to faster, cheaper worker agents. When tested on rebuilding SQLite in Rust, the new swarm achieved 80% test pass rate in four hours using Grok 4.5, significantly outperforming the previous system, while allowing flexible model configurations that trade cost against performance without sacrificing quality.