量子位官网T2·78 pts
- MatOS unifies data, skills, workflows and expert roles into revisable, reusable scientific assets so research capability compounds over time
- It scores 33.6% higher on average than Codex CLI with the same base model on ResearchClawBench, and 80.8% higher on materials tasks; full-task success on MatTools reaches 97.96%
- A hierarchical screening pipeline cut over 30 million candidates to 97 solid electrolytes, and supported DuoCompNet in lowering alloy melting-point MAE to 3.18°C