Accelerating agentic RL and evaluation research velocity with 45x faster GKE Agent Sandbox

Google Cloud released GKE Agent Sandbox and an RL orchestration SDK, generally available today. The update cuts sandbox cold-start times from 45 to 85 seconds down to 1 to 9 seconds. It also reduces worst-case wait times from 7.5 minutes to under 10 seconds. This infrastructure change aims to keep GPU clusters active during parallel agentic reinforcement learning rollouts.