Google made Pathways on Cloud generally available on Google Kubernetes Engine on November 9, 2026. This runtime, used internally to train models such as Gemini, allows a single JAX client to orchestrate workloads across hundreds or thousands of TPU chips. Pathways supports batch training, interactive inference, and multi-host inference, with Llama 3.1 405B cited as an example. The platform includes the PathwaysJob API, support for creating Pathways-enabled clusters through XPK or gcloud, and suspend-resume resiliency and elastic training features.
Source: Read the original article


