Runpod Announces Batch Inference Availability, Adds GLM-5.3 Endpoint
Runpod has announced that batch inference is now generally available on its platform. The announcement page also points to a GLM-5.3 public endpoint, rebuilt PyTorch templates and a guide to Global Volumes.
- Runpod says batch inference is now generally available.
- GLM-5.3 is available on Runpod's public endpoints, usable via Playground, curl or the OpenAI SDK in Python.
- PyTorch templates now cover 2.6 through 2.13 across multiple CUDA and Ubuntu versions.
- The Pod deploy flow has a new CUDA dropdown for matching templates to GPU and workload.
- Global Volumes mount like network volumes but run on object storage.
Batch inference reaches general availability
The post is titled "Batch Inference Is Now Generally Available on Runpod". The text available for this report contains only short summaries of related posts and gives no details on how batch inference works, how it is priced or how it is used. Those specifics are not covered here, and readers should check Runpod's own post for them.
What the page does show is a set of other updates published alongside the announcement, aimed at developers who run models and workloads on Runpod's infrastructure.
GLM-5.3 on public endpoints
According to Runpod, GLM-5.3 is now available on its public endpoints. The company has published a walkthrough with three ways to use it:
- Trying the model in the Playground
- Calling it with curl
- Using it through the OpenAI SDK in Python
Support for the OpenAI SDK means developers can reach the model with a client library they may already use, rather than a separate toolkit. The source does not give pricing or performance figures for the endpoint.
Rebuilt PyTorch templates and Global Volumes
Runpod also says it has rebuilt its PyTorch templates. They now cover versions 2.6 through 2.13 across multiple CUDA and Ubuntu versions. The Pod deploy flow has a new CUDA dropdown, which the company says lets users match a template to their GPU and workload.
A third post explains Global Volumes. Runpod says they mount like a network volume but run on object storage. The company stresses that Global Volumes are not "network volumes, but better", and that the object storage basis changes what users should put on them. The excerpt does not say which kinds of data suit them best.
What it means for users
For Runpod customers, the announcements point to a wider set of options for running inference and managing environments on the platform. Batch inference being generally available suggests the feature is ready for regular use, though the available text does not describe its capabilities. The GLM-5.3 endpoint, the broader range of PyTorch and CUDA combinations, and the guidance on Global Volumes are practical changes for anyone choosing models, templates or storage on Runpod.
RunPod: Batch Inference Is Now Generally Available on Runpod
