Baseten on Hugging Face Inference Providers 🔥

The announcement marks Baseten’s entry into the Hugging Face Inference Providers program, which previously included only a handful of vendors. By being added, Baseten’s models become discoverable alongside Hugging Face’s own offerings and can be selected directly from a model card. This changes the workflow for engineers who previously had to manage separate API endpoints or custom wrappers to use Baseten.

Adopting the provider does not introduce extra fees; when a request is routed through Hugging Face, the Hub passes through the provider’s base cost and applies any Hugging Face-specific credits or limits. If a developer supplies their own Baseten key, billing stays with Baseten, preserving existing cost arrangements. The only operational change is the need to manage an HF token for routed calls or a Baseten key for direct calls.

The current rollout is limited to conversational and text-generation inference; other modalities such as image generation or audio are not yet available through Baseten on the Hub. Consequently, engineers whose workloads rely on those modalities must continue to use Baseten’s native API or another provider until support expands. Additionally, the routed mode depends on the Hugging Face rate limits and token policies, which may constrain high-volume usage.

Because only one feed carries the news, there is no external corroboration to verify details such as the exact list of supported models or the timing of future task additions. Engineers should treat the information as preliminary and consult the Baseten documentation or the Hugging Face provider page for the most up-to-date specifications. The lack of multiple sources reduces confidence in any nuanced claims that are not directly stated in the article.