Article

Gradient has introduced Echo-2, a decentralized reinforcement learning platform that significantly reduces AI model training costs. Announced in early 2025, Echo-2 utilizes a global network of unused computing power, cutting expenses from thousands to hundreds of dollars for training a 30-billion-parameter model.

This innovation is expected to democratize AI development, making it accessible to researchers and startups. Echo-2 employs advanced asynchronous RL techniques and a peer-to-peer protocol, enhancing efficiency while maintaining stability across distributed networks.