Baseten Joins Hugging Face as a Serverless Inference Provider
Developers can route supported open-weight language models to Baseten from Hugging Face model pages and client libraries, using either provider credentials or Hugging Face billing.
Hugging Face added Baseten as a supported Inference Provider on August 6, 2026, giving developers another serverless route for running selected open-weight language models from Hub model pages and client libraries. The first phase covers conversational and text-generation tasks, while support for other task types remains planned rather than available.
The integration affects developers who want hosted inference without wiring a separate provider-specific interface into every application. Hugging Face presents the provider choice inside its existing model-discovery and SDK workflow, while Baseten supplies the underlying serving infrastructure for models it supports.
Baseten becomes another route behind the Hub
Hugging Face's Inference Providers layer lets a compatible model page expose more than one hosted backend. Baseten now appears within that system, and users can place providers in a preferred order for model-page widgets and generated code examples.
The launch article names Kimi K3, DeepSeek V4 Flash and GLM-5.2 among the initially available open-weight models. Those examples illustrate the integration's current emphasis on language-model inference. They do not mean every model or every Baseten capability is available through the Hugging Face route. The announcement says additional task support will arrive later, without giving a date.
For Python users, the integration is available in `huggingface_hub` version 1.26.1 or later. JavaScript users can access it through the Hugging Face inference client. The provider layer is also connected to several agent harnesses, including Pi, OpenCode, Hermes Agents and OpenClaw, allowing supported Baseten-hosted models to sit behind tooling that already understands Hugging Face providers.
Developers can choose credentials and billing paths
The integration exposes two operating modes. A developer can supply a Baseten API key, in which case the request goes directly to Baseten and the provider bills the developer's Baseten account. Alternatively, the request can be routed by Hugging Face with a Hugging Face token, and the charge is applied to the user's Hugging Face account.
Hugging Face says routed requests use the provider's standard API rates without an additional markup. That simplifies consolidated billing, but it is a first-party statement from the platform and should be checked against the live account and model pricing before production use. Provider availability, model selection and rates can change independently of the integration announcement.
This split is useful for teams with different procurement needs. Existing Baseten customers can keep provider-level credentials and billing, while teams already managing access through Hugging Face can use the routed path. The model identifier still selects Baseten as the backend, so routing is explicit rather than an invisible claim that all providers are interchangeable.
The practical change is distribution, not a new model
Baseten did not release a new foundation model in this announcement. The change is a distribution and infrastructure integration: supported models can now be discovered and called through an interface that many open-model developers already use. That can reduce setup work and make it easier to compare or switch hosted backends at the application layer.
The initial scope also defines the limitation. Conversational and text generation are supported first; speech, image and other task classes are not confirmed as part of this launch. The article describes future expansion but provides no release schedule, so developers should treat the current provider list and task compatibility as the authoritative boundary for deployment planning.
The next verifiable checkpoint will be the arrival of additional task types or evidence that the integration broadens model availability beyond the launch set. Until then, the strongest reader value is the new choice of serving route and the separation between direct-provider and Hugging Face-routed billing.
Status
Confirmed. Internal confidence is medium because the integration details and billing description come from the joint product announcement hosted by Hugging Face, without independent operational testing in the evidence reviewed for this article.
Sources
Update note: Last reviewed 2026-08-11. We will revise this article if the supported task boundary, SDK requirements, provider routing or billing model materially changes.
Sources
Drafted with AI assistance from source briefs; reviewed for citation completeness and label accuracy.