Gradient launches decentralized RL platform 'Echo-2'

AI infrastructure company Gradient has announced the launch of 'Echo-2', its next-generation decentralized reinforcement learning (RL) platform. The company stated that the era of data scaling is ending, replaced by an era of "Inference Scaling," where models verify their own logic and find solutions through trial and error. To facilitate this, Gradient built 'Lattica', a peer-to-peer protocol that rapidly deploys model weights to hundreds of edge devices. The platform controls numerical precision at the kernel level, ensuring that a consumer GPU in Seoul and an H100 in Virginia can produce bit-identical results. It also maximizes computational efficiency through an asynchronous orchestration layer that separates the learner from the sampling fleet. Echo-2 has already completed performance verification in areas with real financial responsibilities, including high-level reasoning at the Math Olympiad level, smart contract security audits, and autonomous on-chain agents.