Hugging Face announced that Baseten is now a supported Inference Provider on the Hugging Face Hub. The integration lets developers run serverless inference directly from model pages and via client SDKs for JavaScript and Python, using Baseten-hosted models.
What this enables
Baseten is an AI infrastructure platform offering serverless inference and model hosting among other services. Through the initial integration on the Hub, Baseten provides support for conversational and text-generation tasks, exposing a number of open-weight LLMs such as Kimi K3, the latest DeepSeek V4 Flash, and GLM-5.2. Hugging Face says support for additional task types will be added over time.
How it works — UI and SDK usage
On the Hugging Face website, users can set provider API keys in their account settings. If no custom key is provided, requests will be routed through Hugging Face. Users can also order providers by preference; that order is applied to the model page widget and to code snippets.
There are two modes when calling Inference Providers:
- Custom key: calls go directly to the inference provider using the user’s API key for that provider.
- Routed by Hugging Face: no separate provider token is required; requests are billed to the user’s Hugging Face account and routed through the Hub.
Baseten is accessible from the Hugging Face SDKs — Python via huggingface_hub (>= 1.26.1) and JavaScript via @huggingface/inference. Hugging Face provided example code showing how to call the DeepSeek V4 Flash model hosted by Baseten; authentication uses a Hugging Face token and the request is routed to Baseten automatically.
Integration with Agent Harnesses
Hugging Face Inference Providers are integrated into several Agent Harnesses, including Pi, OpenCode, Hermes Agents, and OpenClaw. This allows Baseten-hosted models to be plugged into these tools without extra glue code.
Billing and credits
When you use a provider’s API key (direct requests), billing is handled by that provider — for example, Baseten API key usage is billed to the user’s Baseten account. For routed requests authenticated via the Hugging Face Hub, standard provider API rates are applied and charged through the Hugging Face account; Hugging Face does not add an extra markup. The company notes that revenue-sharing agreements with provider partners may be considered in the future.
A notable user benefit: Hugging Face PRO subscribers receive $2 of Inference credits each month, which can be used across providers. Hugging Face also offers a small free inference quota to signed-in free users; PRO subscribers gain additional features such as ZeroGPU, Spaces Dev Mode, and higher usage limits.
Feedback and next steps
Hugging Face invites user feedback on the integration. The Hub lists the full set of models currently supported via Baseten, and the platform will roll out support for more tasks and models over time.
