Baseten on Hugging Face Inference Providers 🔥

The AI infrastructure landscape continues to evolve, and today marks a significant milestone for developers working within the Hugging Face ecosystem. We are excited to announce that Baseten—the robust platform known for its serverless AI and model training capabilities—is now officially a supported Inference Provider on the Hugging Face Hub.

This integration bridges the gap between Baseten’s high-performance infrastructure and the Hub’s expansive model catalog, allowing developers to deploy and interact with frontier models directly from their favorite model pages.

Expanding the Serverless Ecosystem

Baseten has built a reputation for simplifying the integration of complex AI capabilities into production applications with minimal configuration. By joining the Hugging Face Inference Provider ecosystem, Baseten brings its specialized infrastructure to a broader audience.

Initially, this partnership focuses on conversational and text-generation tasks. Developers can now leverage Baseten’s backend to run popular open-weight Large Language Models (LLMs), including:

• DeepSeek V4 Flash
• Kimi K3
• GLM-5.2

This is just the beginning; support for additional model types and task categories is slated for release in the near future. You can explore the full, growing catalog of Baseten-supported models directly on their Hugging Face profile.

Seamless Integration: How It Works

The integration is designed for flexibility, offering two distinct ways to interact with models:

1. Custom API Keys

For power users who already maintain a Baseten account, you can input your own API keys within your Hugging Face user settings. In this mode, your requests are routed directly to Baseten, and you are billed according to your existing Baseten subscription.

2. Routed via Hugging Face

For those who prefer a unified workflow, you can opt to have your requests routed through Hugging Face. In this scenario, you do not need a separate provider token; the costs are simply applied to your Hugging Face account. This "pass-through" billing ensures there is no additional markup, keeping costs transparent and predictable.

"Inference Providers are seamlessly integrated into our client SDKs, making it incredibly straightforward to switch between models and providers with minimal code changes."

Developer Experience: SDKs and Agent Harnesses

Whether you are working in Python or JavaScript, the integration is built to be plug-and-play. By utilizing the huggingface_hub (v1.26.1 or higher) for Python or the @huggingface/inference library for JavaScript, developers can call Baseten-hosted models with just a few lines of code.

Furthermore, Baseten-hosted models are now compatible with a wide array of Agent Harnesses. Tools such as Pi, OpenCode, Hermes Agents, and OpenClaw can now utilize Baseten’s infrastructure without requiring complex "glue code," significantly accelerating the development of autonomous agents.

Billing and Pro Benefits

Hugging Face continues to incentivize the use of its platform through its Pro subscription model.

• Pro Perks: Subscribers receive $2 in monthly Inference credits, which can be applied across various providers, including Baseten.
• Free Tier: Signed-in users still receive a small quota for free inference, though upgrading to Pro is highly encouraged to unlock higher rate limits, ZeroGPU access, and Spaces Dev Mode.

Looking Ahead

This collaboration represents a major step forward in making frontier AI more accessible. By combining the ease of the Hugging Face Hub with the raw performance of Baseten’s serverless infrastructure, developers have more power than ever to build sophisticated AI applications.

We are eager to hear how the community utilizes these new capabilities. If you have feedback, feature requests, or comments on this integration, please join the conversation on the Hugging Face community discussion board. Stay tuned for more updates as we continue to expand the list of supported tasks and models.