#qwen3

Small Language Models Outperform Large Systems in Tool Selection

A study by Meta and Carnegie Mellon University shows that small, reinforcement-trained language models can outperform massive systems in selecting software tools for AI agents.