Hugging Face has added Baseten as a supported Inference Provider on the Hub, giving developers another serverless option for running models directly from model pages and client SDKs. Baseten is an AI infrastructure platform covering serverless inference, training, and more, with a catalog of frontier models.

The integration initially supports conversational and text-generation tasks, including open-weight LLMs such as Kimi K3, DeepSeek V4 Flash, and GLM-5.2. More task types are planned for future rollouts.

Users can configure Baseten in two ways from the Hub UI: set a custom Baseten API key so requests go directly to Baseten (billed on your Baseten account), or use Hugging Face routed mode where no Baseten key is needed and charges appear on your HF account with no additional markup. PRO subscribers get $2 monthly in Inference credits usable across providers.

The Baseten provider is also accessible through Hugging Face SDKs (huggingface_hub >= 1.26.1 for Python and @huggingface/inference for JavaScript). Examples show using DeepSeek V4 Flash through the OpenAI-compatible endpoint at router.huggingface.co/v1, authenticating with an HF token. Many agent harnesses — including Pi, OpenCode, Hermes Agents, and OpenClaw — already integrate Hugging Face Inference Providers, letting developers plug Baseten-hosted models directly without glue code.