Gradient Launches Decentralized RL Platform 'Echo-2'

AI infrastructure company Gradient announced the launch of its next-generation distributed reinforcement learning (RL) platform, Echo-2. Gradient believes that the era of data expansion is over and the era of inference scaling, where models verify their own logic and find the right answer through trial and error, has arrived. They built Lattica, a P2P protocol that rapidly distributes model weights to hundreds of edge devices, and controlled numerical precision at the kernel level to ensure that consumer GPUs in Seoul and H100s in Virginia produce bit-for-bit identical results. Furthermore, they maximized computational efficiency through an asynchronous orchestration layer that separates the learner and sampling fleets. Echo-2 has completed performance verification in areas with real-world financial responsibility, such as Mathematical Olympiad-level difficult inference, smart contract security audits, and autonomous on-chain agents.