Gradient Network Information

Gradient is an AI research lab building open infrastructure for the future of AI. We believe the most important AI capabilities should not be locked inside a handful of data centers. Our work focuses on making reinforcement learning, model serving, and distributed compute accessible at scale. Our Open Intelligence Stack: Parallax is a distributed inference engine that turns heterogeneous devices into a unified AI cluster. It handles automatic model sharding, dynamic device management, and intelligent request routing across mixed hardware, from NVIDIA GPUs to Apple Silicon. Parallax powers local, offline AI deployment without performance compromise. Echo is our distributed reinforcement learning framework. Echo-2 decouples rollouts from training, enabling asynchronous RL across heterogeneous GPU pools. In benchmarks, it reduced post-training costs for a 30B parameter model from approximately $4,490 to $425, a 90%+ reduction. Logits, our RL-as-a-Service platform built on Echo-2, is in development. Lattica is our peer-to-peer data communication protocol. It provides the networking layer that connects distributed nodes across the Gradient architecture, enabling real-time coordination at global scale. Lattica already runs across millions of peers worldwide. Gradient is backed by a $10M seed round by Pantera Capital, Multicoin Capital, and HSG (prev. Sequoia China). We are hiring across research and engineering. See open roles at careers.gradient.network