Echoverse: Deep, evolving environments for computer-use agents
Microsoft Research Akshay Nambi, Yash Pandya, Sahil Gupta, Sarthak Harne, Archana Yadav, Kavyansh Chourasia, Yash Lara, Ahmed Awadallah, Ece Kamar
Microsoft released Echoverse, a collection of twelve training environments for computer-use AI agents, featuring ten deep domain worlds and two capability worlds that simulate real applications with stateful behavior and realistic data. A 9-billion-parameter model trained on all twelve worlds improved from 36.5% to 67.1% accuracy, approaching GPT-5.4's performance within fourteen percentage points. The approach emphasizes co-evolving the environments, tasks, and verification systems alongside model training rather than simply scaling the number of shallow simulation worlds.
Why it matters
Computer-use AI agents struggle with multi-step workflows like email and customer support. Echoverse trains agents in realistic environments rather than simply providing more training tasks, helping them improve as the tasks, tests, and environments evolve. The post Echoverse: Deep, evolving environments for computer-use agents appeared first on Microsoft Research.
