AI news · October 18, 2026
RunPod adds batch jobs and location-free volumes
RunPod batch jobs accept up to 5,000 requests per batch in public beta; use them for offline scoring instead of maintaining your own queue.
- requests per batch
- 5,000
- active batches per endpoint
- 10
- queued requests in beta
- 50,000
RunPod's Q3 2026 release report adds a public beta for Serverless Batch Jobs. A batch can hold up to 5,000 requests, and each endpoint can run up to 10 active batches with 50,000 requests queued during the beta. RunPod also added worker affinity for workloads that keep a KV cache, six endpoint creation paths, and REST API v2 for Pods, Serverless, Clusters, templates, storage, credentials, hardware, and billing.
Global Volumes are also in public beta and can be attached to Pods or Serverless endpoints without choosing a data center first. RunPod says they fit model weights, LoRAs, tokenizers, and configs that are read often and written rarely; storage and file operations are billed separately. The release is useful for batch scoring and shared model storage, but a small workload should confirm queue time and the full bill before a move.
What you can do with it
Take one offline scoring job and split it into a fixed batch. Compare your current client-side queue with RunPod Batch Jobs on request count, wait time, retries, completed outputs, and storage cost. Keep the beta path only if the simpler queue handling survives a repeat run.
Our take
The strongest release is the batch interface because it removes a repeated piece of application code. Global Volumes can help when the best GPU is not near the current file store, but separate storage charges matter. Treat the first run as an operations check, not a promise of lower cost.
Links Read the release
Source: RunPod — Made With Models writes the brief; the reporting is theirs.