Baseten joins Hugging Face Inference Providers with support for open-weight LLMs

Baseten, an AI infrastructure platform offering serverless AI and training, announced its integration as a Hugging Face Inference Provider. The partnership enables developers to run open-weight models directly on Hugging Face model pages or via the Hugging Face SDKs for Python and JavaScript.

Supported models include DeepSeek V4 Flash, Kimi K3, and GLM-5.2, with additional tasks beyond conversational and text-generation slated for future updates. The integration supports both direct and routed billing: developers can use their own Baseten API key for direct billing or rely on Hugging Face’s routing, which bills through the user’s Hugging Face account without markup.

The integration is accessible through Hugging Face’s client SDKs (huggingface_hub for Python and @huggingface/inference for JavaScript) and is also compatible with several agent harnesses, including Pi, OpenCode, Hermes Agents, and OpenClaw. This allows developers to plug Baseten-hosted models into their workflows with minimal configuration.

Hugging Face notes that free-tier users have a small inference quota, while PRO users receive $2 worth of monthly inference credits usable across providers. Future revenue-sharing agreements with providers are mentioned as a possibility.