Baseten on Hugging Face Inference Providers 🔥

Baseten Joins Hugging Face Inference Ecosystem

Hugging Face has expanded its platform capabilities by adding Baseten, an AI infrastructure provider, to its official list of supported Inference Providers. The integration allows developers to run models on Baseten's serverless infrastructure directly from Hugging Face model pages and through its client SDKs. This partnership aims to streamline the process of deploying and scaling a variety of open-weight models by abstracting away the underlying infrastructure management for developers.

Technical and Billing Integration

The integration provides two primary methods for accessing models. Users can either supply their own Baseten API key for direct billing and access, or they can route requests through Hugging Face using an HF token, with charges applied to their Hugging Face account. According to the announcement, Hugging Face will pass through the provider costs directly without any additional markup. The initial rollout focuses on conversational and text-generation tasks, with more to follow.

• Initial Models: Support includes popular LLMs such as Kimi K3, the latest DeepSeek V4 Flash, and GLM-5.2.
• Dual Access Modes: Developers can use a direct Baseten API key for billing on their Baseten account or a Hugging Face token for unified billing through the HF platform.
• SDK Compatibility: The integration is available in `huggingface_hub` for Python (version 1.26.1 and higher) and `@huggingface/inference` for JavaScript.
• Agent Harnesses: Baseten-hosted models can be used directly within agent frameworks like Pi, OpenCode, and Hermes Agents without custom code.

Ecosystem Implications

By incorporating providers like Baseten, Hugging Face continues to solidify its position as a central hub for the entire AI development lifecycle, moving beyond being just a model repository. This creates a competitive marketplace for inference providers, giving them direct access to a large developer community. For developers, this lowers the barrier to experimenting with and deploying a diverse set of high-performance models, increasing optionality and potentially driving down costs as providers compete on performance and price within a single, unified interface.

This integration further establishes Hugging Face as a neutral aggregator and marketplace for AI compute, where infrastructure providers like Baseten compete for developer workloads. It signals a market maturation where developer experience and ease of access are becoming as important as the underlying model performance itself.