Baseten announced that its model inference service is now integrated into Hugging Face's Inference Providers, allowing users to access Baseten's inference capabilities directly through the Hugging Face platform. This collaboration simplifies the model deployment process and offers developers more options.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
Hugging Face has announced Baseten as a new member of its Inference Providers ecosystem, allowing developers to call open-weight models hosted by Baseten directly from the Hub or via SDKs. This integration highlights HF's dual-track design of 'no-markup resale' and 'custom key direct connection' for inference distribution, along with a $2 monthly inference credit incentive for PRO users.
The Hugging Face official blog announced on August 6, 2026, that Baseten has become a new member of its Inference Providers ecosystem. Baseten is described as an AI infrastructure platform covering serverless AI, training, and more, with a catalog of cutting-edge models, allowing developers to integrate various AI capabilities with minimal setup.
As part of the initial integration, Baseten supports chat and text generation tasks on Hugging Face, with access to open-weight models such as Kimi K3, the latest DeepSeek V4 Flash, and GLM-5.2. The official announcement states that support for more task types will be coming soon.
In the user account settings on the Hugging Face website, users can set their own API keys and also sort providers by preference, which affects the components and code snippets on model pages. There are two modes for calling Inference Providers: custom key mode (requests are sent directly to the inference provider using the user's own API key) and HF routing mode (users do not need a provider token, and costs are billed directly to the HF account).
This dual-track design means developers can choose to connect directly or route through HF based on their needs. For routing mode, HF explicitly states that it does not charge an additional markup; it only passes through provider costs, but revenue-sharing agreements may be established in the future.
Baseten is now available through Hugging Face SDKs: Python's huggingface_hub (>= 1.26.1) and JavaScript's @huggingface/inference. The blog provides example code using an OpenAI-compatible interface, setting base_url to https://router.huggingface.co/v1 and using an HF token to call the model deepseek-ai/DeepSeek-V4-Flash-0731:baseten.
Additionally, Inference Providers have been integrated into most agent frameworks, including Pi, OpenCode, Hermes Agents, OpenClaw, and others, meaning users can plug Baseten-hosted models directly into these tools without additional glue code.
In terms of billing, direct requests (using provider keys) are billed by the respective provider; for example, using a Baseten API key charges the Baseten account. Routed requests (via HF authentication) are paid at standard provider API rates, with no markup from HF.
HF PRO users receive $2 in inference credits monthly, usable across all providers; free logged-in users also get a small amount of free inference quota. These incentives aim to attract users to try new providers.
All information in this article comes from the Hugging Face official blog, which is a first-party publication and thus highly credible. However, the model names mentioned (such as Kimi K3, DeepSeek V4 Flash, GLM-5.2) and future plans (such as more task support and revenue sharing) are official statements and have not been independently verified.
The addition of Baseten further enriches Hugging Face's inference provider ecosystem. Its dual-track calling modes and transparent billing strategy lower the barrier for developers, while the monthly credit for PRO users may drive broader adoption.
Primary report
Primary source