7/27/2026
•
EN
Evaluating Agents Beyond the First Prompt
EvoCode-Bench is a multi-turn coding benchmark that tests AI agents on evolving tasks across 227 rounds, evaluating persistence and adaptability.