TL;DR
Baseten is now available through Hugging Face Inference Providers for conversational and text-generation workloads. Developers can route requests through Hugging Face or use a Baseten API key, but performance, availability and the schedule for more task types remain unspecified.
Hugging Face has added Baseten as a supported Inference Provider, allowing developers to send conversational and text-generation requests to Baseten-hosted models from the Hugging Face Hub and compatible software. The integration adds another infrastructure option for accessing open-weight language models without building a separate connection to each model-serving platform.
The initial release covers chat and text generation. Hugging Face identified Kimi K3, DeepSeek V4 Flash and GLM-5.2 among the models available through Baseten, while directing users to Baseten’s Hub profile for the current catalog. These model names and their availability reflect Hugging Face’s announcement; the company did not publish independent performance comparisons in the post.
Users can access Baseten through two billing and authentication paths. A developer who supplies a Baseten API key sends requests directly to the provider and receives charges on the Baseten account. A developer using a Hugging Face token can have the request routed through the Hugging Face infrastructure, with provider costs charged to the user’s Hugging Face account.
Baseten is also exposed through huggingface_hub version 1.26.1 or later for Python and through @huggingface/inference for JavaScript. Hugging Face said its provider router works with an OpenAI-compatible chat-completions interface, and named Pi, OpenCode, Hermes Agents and OpenClaw among the agent tools that can use Inference Providers.
Baseten joins Hugging Face Inference Providers
Developers can now route conversational and text-generation workloads to Baseten-hosted models from Hugging Face Hub pages, supported SDKs and compatible agent tools.
One model interface. Two billing paths. More serving choice.
The launch expands infrastructure options for open-weight models, while performance benchmarks and the timetable for additional tasks remain unspecified.
Python SDK ≥ 1.26.1Baseten is now selectable inside the Hugging Face ecosystem
The first release focuses on language-model inference. Availability is surfaced through compatible model pages, client libraries and the Hugging Face provider router.
Conversational
OpenAI-compatible chat-completions requests can be routed to supported Baseten-hosted models.
Text generation
Developers can generate text through Hub integrations without writing a separate provider-specific connection.
Initial model set
Hugging Face highlighted Kimi K3, DeepSeek V4 Flash and GLM-5.2. The live catalog remains the current reference.
How a routed request moves
Teams select Baseten in a supported model identifier while retaining a familiar Hugging Face entry point.
Choose a model
Start on a supported Hub page or specify the model in code.
Select Baseten
Use provider ordering or identify Baseten in the request.
Authenticate
Supply either a Hugging Face token or a Baseten API key.
Run inference
The supported workload is served through Baseten infrastructure.
Two ways to authenticate and pay
Architecture and billing ownership depend on which credential is supplied with the request.
Hugging Face token
Hugging Face routes the request and charges the provider cost to the developer’s Hugging Face account.
Baseten API key
The request uses the developer’s provider credential, with usage charged directly to the Baseten account.
Where developers can use it
The integration spans official client libraries, model-page experiences and software using the compatible chat-completions interface.
| Access surface | Baseten support | Credential choice | Key detail |
|---|---|---|---|
| Hugging Face Hub pages | ✓ Available | HF token or provider key | Provider appears on compatible models |
| Python: huggingface_hub | ✓ Available | HF token or Baseten key | Version 1.26.1 or later |
| JavaScript: @huggingface/inference | ✓ Available | HF token or Baseten key | Supported inference client |
| OpenAI-compatible agent tools | ✓ Compatible | Typically HF router token | Pi, OpenCode, Hermes Agents, OpenClaw |
| Tasks beyond chat and text | ~ Planned | Not specified | No published rollout calendar |
| Published performance comparison | ✗ Not provided | Not applicable | Production teams must benchmark directly |
What the announcement says—and what it does not
The integration is confirmed, but the launch post is not an independent performance study or a production service-level guarantee.
“Baseten is now a supported Inference Provider on the Hugging Face Hub.”
Hugging Face announcement“There’s no additional markup from us; we just pass through the provider costs directly.”
Hugging Face billing statementStill unspecified
- Latency, throughput and reliability comparisons
- Regional availability and capacity constraints
- Model-by-model service levels
- Sequence and release dates for additional task types
- Long-term pricing or commercial arrangements
Confirmed capability versus missing detail
These bars visualize disclosure completeness in the announcement—not measured product performance.
Bottom line for production teams
Baseten adds meaningful routing choice and a flexible billing decision. Before deployment, verify the live model catalog, current pricing, quotas, regional coverage and workload-specific latency and reliability.
More Choice for Model Routing
The addition gives Hugging Face users another serving provider without requiring them to leave a model page or create provider-specific application logic. Teams can arrange providers by preference and use the same Hugging Face routing endpoint while selecting Baseten in the model identifier. That may make it easier to compare infrastructure options or change providers when a model is supported by more than one service.
The billing paths also give teams a choice between centralized Hugging Face charges and a direct Baseten relationship. Hugging Face said routed requests carry the provider’s standard API rates with no added markup. That statement describes the platform’s current billing policy, not a guarantee that provider prices or commercial arrangements will remain unchanged.
As an affiliate, we earn on qualifying purchases.
How Hugging Face Routing Works
Inference Providers is Hugging Face’s system for connecting model pages and client libraries to third-party inference services. Account settings allow users to save provider keys and rank providers by preference. Compatible providers then appear in model-page widgets and code examples, based on the selected model and the user’s ordering.
Baseten describes itself as an AI infrastructure platform spanning serverless inference, model training and other deployment services. Hugging Face said the platform supports model categories ranging from large language models to text-to-speech systems, although the first Hugging Face release is limited to conversational and text-generation tasks.
Hugging Face also offers $2 in monthly inference credits to PRO subscribers for use across providers. Signed-in free users receive a small free inference quota, according to the announcement. Actual usage depends on model pricing, token volume and the provider selected for a request.
“Baseten is now a supported Inference Provider on the Hugging Face Hub.”
— Hugging Face
As an affiliate, we earn on qualifying purchases.
Performance and Rollout Details Missing
Hugging Face did not provide latency, throughput or reliability measurements for Baseten-backed requests, so the announcement does not establish how the service compares with other providers. It also did not specify regional availability, capacity limits or model-by-model service levels. Developers evaluating production use will need current documentation and their own workload tests.
The companies have not given a date for supporting more inference tasks. The announcement says more task types will arrive soon, but does not identify which capabilities will be added first. The precise model catalog may also change, and pricing remains provider-dependent.
As an affiliate, we earn on qualifying purchases.
More Baseten Tasks Planned
Hugging Face and Baseten are expected to expand the supported task list beyond chat and text generation, though no release calendar has been published. Near-term signals will include changes to Baseten’s supported-model catalog, updates to the SDK documentation and new task options appearing on compatible Hub pages.
Developers can now test the integration by selecting Baseten on a supported model page or by making authenticated requests through the Hugging Face router. Production users will need to check current model availability, pricing and usage limits before choosing between direct Baseten billing and Hugging Face-routed access.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did Hugging Face announce about Baseten?
Baseten is now a supported Hugging Face Inference Provider. The first release lets users run conversational and text-generation models through Baseten from supported Hub pages, client libraries and compatible agent software.
Which models are available through the integration?
Hugging Face named Kimi K3, DeepSeek V4 Flash and GLM-5.2 as examples. The live Baseten catalog on Hugging Face is the better reference for current availability because supported models can change.
Do users need a Baseten API key?
No Baseten key is required when a request is authenticated with a Hugging Face token and routed through Hugging Face. Users can instead provide their own Baseten key for direct requests billed to a Baseten account.
Does Hugging Face add a fee to routed requests?
Hugging Face said it currently passes through standard provider API rates without an added markup. Charges depend on the selected model and provider pricing, which may change over time.
When will Baseten support more task types?
No rollout date has been announced. Hugging Face said more tasks are planned, but the sequence and scope remain unspecified.
Source: Hugging Face
Source: Hugging Face
