A collaborative agent with two lightweight synergistic models for autonomous crystal materials research

TL;DR - Shi et al. present a collaborative agent that combines two lightweight large language models for autonomous crystal materials research. The architecture aims to make capable scientific reasoning and tool use affordable enough for local deployment.

• Uses a dual-model architecture in which lightweight LLMs collaborate on research tasks.
• Integrates scientific reasoning with execution of external research tools.
• Achieves competitive performance without relying on larger, more costly models.
• Supports accessible, locally deployable automation for crystal materials research.