DeepInfra added to Hugging Face Inference Providers list

DeepInfra has been added to Hugging Face Inference Providers, making DeepInfra a selectable backend for running models published on the Hugging Face Hub. The integration lets Hub users route inference requests to DeepInfra endpoints and use DeepInfra billing and compute when they deploy models through Hugging Face tooling.

The addition appears in the Hugging Face Inference Providers list exposed in the Hub and the Hugging Face Inference API. Users can now choose DeepInfra as the execution provider for model inference from the Hub interface or via provider selection in API calls, with authentication and provider-specific billing handled by DeepInfra.

How the integration works

When a developer selects DeepInfra as the inference provider, calls to a model on the Hugging Face Hub are forwarded to DeepInfra's infrastructure rather than running on Hugging Face-hosted compute. The Hub surfaces provider options alongside model pages and the Inference API. Users will need to link or supply DeepInfra credentials where required and accept DeepInfra's pricing and terms for any compute consumed through that provider.

The integration does not move models off the Hugging Face Hub. Models remain hosted in the Hub and access controls such as repository visibility and access tokens continue to govern who can invoke a model. The provider selection changes only where the inference workload runs and which organization invoices for it.

What developers should expect

Developers choosing DeepInfra should expect a provider-specific endpoint and metrics shown through DeepInfra's dashboard as well as whatever usage summaries Hugging Face surfaces for third-party providers. Latency, throughput, model startup behavior and supported accelerator types will depend on DeepInfra's offering and the instance types it exposes to Hugging Face users.

From a workflow perspective the integration is intended to be plug and play: select DeepInfra when configuring an inference call in the Hub UI or include the provider parameter in the Inference API request. Error handling, retries and model compatibility remain responsibilities of the provider and the model maintainer. Teams that rely on particular runtime formats or accelerator types should verify compatibility and performance on DeepInfra before migrating production traffic.

Hugging Face's list of providers gives model authors and application developers more choices for where their code executes. The provider model also keeps billing and support with the provider, so invoice and SLA questions are handled outside Hugging Face when DeepInfra is selected.

Why it matters

The addition of DeepInfra increases choice for teams deploying models from the Hugging Face Hub, offering another path for GPU-backed inference and provider-level billing. This raises pressure on pricing and performance among inference vendors and gives developers more options to match cost and latency needs to a provider. Organizations that need specific instance types or vendor relationships now have another integrated option to evaluate when deploying Hub models.