Glonce

Baseten joins Hugging Face Inference Providers

infrastructureThe Glonce editorsHugging Face🔥53Glonce rating53Impact45Novelty35Relevance88Surprise25Players50Credibility88
Baseten joins Hugging Face Inference Providers

Baseten has been added as a supported Inference Provider on the Hugging Face Hub. Baseten is an AI infrastructure platform covering serverless inference, training and more, and it joins Hugging Face's growing roster of inference providers that appear directly on model pages and inside the official client SDKs. As part of this initial integration, Baseten is launching support for conversational and text-generation tasks on Hugging Face, giving access to open-weight LLMs such as Kimi K3, the latest DeepSeek V4 Flash, and GLM-5.2, among others. Hugging Face says support for additional task types will roll out soon, though the post does not specify which ones. Developers can reach Baseten-hosted models three ways: through the model page UI, through the client SDKs (huggingface_hub version 1.26.1 or newer for Python, and the @huggingface/inference package for JavaScript), or through agent harnesses that already integrate Hugging Face Inference Providers, including Pi, OpenCode, Hermes Agents and OpenClaw. Two billing modes are available. With a custom API key, requests go directly to Baseten and are billed on the user's Baseten account. With requests routed through Hugging Face, the user authenticates with an HF token instead, pays the same provider rates with no added markup, and Hugging Face says it may set up revenue-sharing agreements with provider partners in the future. Hugging Face PRO subscribers get $2 worth of Inference credits every month, usable across any provider, on top of the plan's 20x higher rate limits, ZeroGPU access and Spaces Dev Mode. Free signed-in users get a small inference quota as well. Hugging Face is inviting feedback on the integration through a dedicated discussion thread.

Key facts

  • Baseten is now a supported Inference Provider on the Hugging Face Hub, covering serverless inference, training and more.
  • Initial support covers conversational and text-generation tasks, with access to open-weight models including Kimi K3, DeepSeek V4 Flash and GLM-5.2.
  • Access is available via the Hub UI, the client SDKs (huggingface_hub >= 1.26.1 for Python, @huggingface/inference for JS), and agent harnesses such as Pi, OpenCode, Hermes Agents and OpenClaw.
  • Billing is either direct (via a Baseten API key, billed by Baseten) or routed through Hugging Face at the same provider rates with no markup.
  • Hugging Face PRO subscribers get $2 of monthly Inference credits usable across providers, plus 20x higher rate limits.

Why it matters

This extends Hugging Face's Inference Providers program, which lets developers call third-party inference backends directly from the Hub without setting up separate infrastructure for each one. Adding Baseten broadens the pool of open-weight models available this way and reduces the setup work needed to try a new provider.

Who it affects

Developers already building on the Hugging Face Hub or its SDKs, particularly anyone using open-weight LLMs like Kimi K3, DeepSeek V4 Flash or GLM-5.2, and users of agent harnesses such as Pi, OpenCode, Hermes Agents and OpenClaw that already plug into Hugging Face Inference Providers.

How to use it

Models can be called from a model page's widget, via the Python or JavaScript SDKs (huggingface_hub 1.26.1+ or @huggingface/inference), or through a supported agent harness, with the OpenAI-compatible endpoint at router.huggingface.co. Users choose between a direct Baseten API key (billed by Baseten) or routing through their Hugging Face account and token (billed at the same provider rate, no markup). PRO subscribers get $2 of monthly Inference credits usable across any provider; free signed-in users get a small quota.

How solid is it

This is a first-party announcement from Hugging Face's own blog describing a shipped integration, not a rumor or forecast. The post includes working code samples for both Python and JavaScript, which is a concrete signal the integration is live rather than announced-but-pending.

Risks and caveats

Hugging Face says support for task types beyond conversational and text-generation will roll out soon but does not say which ones or when. The post does not name an individual author, does not date the announcement, and does not enumerate the full list of models Baseten supports, pointing instead to an external catalog page.