Baseten joins Hugging Face Inference Providers, expanding serverless AI access

Baseten joins Hugging Face Inference Providers

Baseten, an AI infrastructure platform offering serverless inference and model training, is now available as a supported Inference Provider on the Hugging Face Hub. According to the Hugging Face Blog, the integration allows developers to access conversational and text-generation models—including open-weights LLMs such as DeepSeek V4 Flash, Kimi K3, and GLM-5.2—directly from Hugging Face model pages without vendor lock-in or additional configuration.

The move expands Hugging Face’s growing ecosystem of third-party inference providers, which are surfaced on model cards and integrated into Hugging Face’s Python and JavaScript SDKs (huggingface_hub ≥ 1.26.1 and @huggingface/inference). Developers can now select Baseten as their preferred inference backend when working with compatible models.

Two billing and authentication paths

Hugging Face Inference Providers support two operational modes. Developers can route requests through Hugging Face directly—no Baseten account required—with charges applied to their Hugging Face account balance. Alternatively, users can provide their own Baseten API key in account settings for direct API calls, billing against their Baseten subscription. Users can also order provider preferences in the settings, which determines the default provider displayed on model pages and SDK code snippets.

Ecosystem integration with agent frameworks

According to the Hugging Face Blog, Baseten models are pre-integrated into most major AI agent harnesses, including Pi, OpenCode, Hermes Agents, and OpenClaw. This means developers can plug Baseten-hosted models into agentic workflows without writing provider-specific glue code.

Why This Matters

For teams evaluating serverless inference providers, this integration reduces friction in model selection and deployment. Developers building on Hugging Face no longer need to choose between the Hub’s native inference and separate provider accounts—Baseten-hosted models now appear as an option alongside other providers in the same workflow. The dual billing model (account-routed or API-key-direct) appeals to both individual developers and enterprises with existing Baseten commitments.

Support for additional task types beyond conversational and text-generation is promised, which could broaden Baseten’s utility for multimodal and code-generation use cases. For Hugging Face users already familiar with the Hub’s SDK and agent integrations, the frictionless addition of a new provider strengthens the platform’s position as a neutral hub rather than a locked-in inference service.