Baseten has joined Hugging Face as a supported inference provider, bringing serverless model hosting directly to the platform's repository, Hugging Face announced on August 6, 2026.
The initial integration introduces support for conversational and text-generation workloads across open-weight models, including Kimi K3, DeepSeek V4 Flash, and GLM-5.2. Developers can access Baseten's catalog through official Hugging Face software development kits, specifically huggingface_hub version 1.26.1 or later in Python and the @huggingface/inference library for JavaScript. Hugging Face stated that support for additional machine learning tasks will roll out in subsequent updates.
Calls to Baseten operate through two distinct configurations on the platform. Users can supply their own Baseten API key to direct calls to the provider, which bills the external Baseten account. Alternatively, requests can be routed through Hugging Face using a standard platform token, consolidating charges onto the user's Hugging Face account at base provider rates with no added markup.
Third-party agent harnesses, such as Pi, OpenCode, Hermes Agents, and OpenClaw, can connect directly to Baseten-hosted models through the router endpoint without additional adapter code. Account settings on the platform allow users to establish provider preferences, which determine how third-party services appear across model widgets and sample code snippets.
Hugging Face Pro subscribers receive two dollars in monthly inference credits that apply across all supported providers, alongside ZeroGPU access, Spaces developer mode, and twenty-fold limit increases. Free registered accounts retain access through a smaller baseline inference quota, while Hugging Face noted it may explore revenue-sharing arrangements with provider partners in the future.
