Baseten Joins Hugging Face as an Inference Provider, Bringing Popular AI Models to More Developers
Developers can now run open-weight AI models through Baseten directly from the Hugging Face Hub, with no markup on costs and two billing paths to choose from.
Key points
- Hugging Face added Baseten as a supported Inference Provider on its model Hub in 2025.
- Baseten initially supports conversational and text-generation tasks, including models such as Kimi K3 and DeepSeek V4 Flash.
- Hugging Face PRO subscribers receive $2 in monthly inference credits usable across all providers, including Baseten.
- Developers authenticate with a single Hugging Face token; Hugging Face routes the request to Baseten automatically.
- Hugging Face charges no markup on routed requests, passing provider costs through at standard rates.
Hugging Face, the platform where researchers and companies share AI models, has added Baseten to its list of Inference Providers. An Inference Provider is a service that actually runs an AI model and returns a result. Before this change, developers who wanted Baseten-hosted models had to set that up separately. Now they can do it straight from the Hugging Face model page.
Baseten is an AI infrastructure platform that handles the server work so a developer can call a large language model (the technology behind chatbots such as ChatGPT) with a few lines of code instead of managing their own computers.
What models are available right now?
At launch, Baseten supports text-generation and conversational tasks on Hugging Face, covering open-weight models whose underlying design is publicly available. The initial lineup is Kimi K3, DeepSeek V4 Flash, and GLM-5.2. We reported on GLM-5.2's safety record on 4 August when the model was evaluated against Western systems. More task types are coming.
| Model | Type | Status |
|---|---|---|
| Kimi K3 | Open-weight LLM | Available now |
| DeepSeek V4 Flash | Open-weight LLM | Available now |
| GLM-5.2 | Open-weight LLM | Available now |
| Additional task types | Text-to-speech, others | Rolling out soon |
How does billing work?
Two paths exist. Supply your own Baseten API key (a private code that identifies your account) and charges go directly to your Baseten account. Authenticate through Hugging Face instead and the cost appears on your Hugging Face bill at the standard provider rate, with no markup added.
PRO subscribers get $2 in inference credits each month, usable across any supported provider. Free, signed-in users get a small quota at no charge; past that limit, upgrading to PRO unlocks the credits and higher usage caps.
Should ordinary users care about this?
If you build apps or experiment with AI tools, it matters: one account, one token, access to multiple providers without juggling separate API keys. For people who just use finished apps the change is invisible, but it may quietly improve reliability since developers can switch providers without rewriting their code.
The integration also works inside AI agent tools, software that carries out multi-step tasks autonomously, including Pi, OpenCode, and Hermes Agents among others, so models running on Baseten slot into those workflows without extra setup.
Common questions
Do I need a Baseten account to try this?
No. Authenticate with your Hugging Face token and the request routes through Hugging Face. A Baseten account is only needed if you want to bill directly to it using your own Baseten API key.
Is the Hugging Face Python or JavaScript library compatible?
Yes. The huggingface_hub Python library at version 1.26.1 or higher and the @huggingface/inference JavaScript package both support Baseten as of this launch.
