MatOS preprint reports gains in multistage materials research

Researchers report in a preprint that MatOS, an AI system for coordinating materials research, improved on the previous best-performing agent by 11.15% on ResearchClawBench using the same base model. MatOS organizes research data and tools as reusable assets, runs workflows with scientific checks that allow tasks to resume and be traced, and revises assets using execution records and user feedback. Across eight research cases, the system narrowed more than 30 million structures to 97 solid-electrolyte candidates and identified 89 halide variants with higher computed conductivity than their parent structures. Those conductivity results are computational findings.