Together AI announces the public preview of preemptible compute for Together GPU Clusters on Kubernetes, offering a lower-cost option for interruption-tolerant workloads. Preemptible nodes utilize unused NVIDIA accelerated capacity and are billed at a flat 50% of the on-demand rate.

  • Nodes can be reclaimed when capacity is needed elsewhere, triggering a five-minute drain sequence where pods receive SIGTERM to checkpoint and exit.
  • Billing is sub-hourly, with usage metered every one to two minutes.
  • Preemptible nodes join existing clusters via the label, allowing critical components to remain on standard nodes.
  • The feature supports short experiments, ablations, and temporary inference bursts that can resume from checkpoints or retry after interruption.

This allows teams to leverage spare capacity for non-critical workloads like batch jobs and config sweeps while keeping user-facing replicas on guaranteed standard infrastructure.