# Google engineers explain how OpenRL runs reinforcement learning on Kubernetes

> Original title: Google’s GKE Labs turns reinforcement learning into a Kubernetes workload

- Source: SiliconANGLE · AI
- Published: 2026-11-12T01:51:03.000Z
- Bad Influence: https://badinfluence.news/items/m3tm3jxp54i2g78f6wmuflq6r
- Original: https://siliconangle.com/2026/11/11/model-fine-tuning-kubernetes-google-s-openrl-explained-kubeconna/

## Summary

Google engineers Sunil Arora and Justin Santa Barbara explain how OpenRL separates the reinforcement learning loop from compute infrastructure on Kubernetes. Arora says OpenRL, a self-hosted training API released as a research preview in June 2026, lets researchers focus on their RL loop while infrastructure engineers handle scaling. Google's Agent Substrate packs more sandboxes onto limited compute for reward calculation, according to Arora.