Inference Endpoints
Deploys a model from the Hub to a dedicated managed endpoint, billed by the hour.
Find alternatives to Inference Endpoints
What it does
- Model inference
Features 5
- AutoscalingScales an endpoint up as traffic increases and down as it decreases to save compute cost.
- Logs and metrics observabilityLogs and metrics to understand and debug what is happening in the model.
- Open-source inference engine supportDeploys models with open-source inference engines such as vLLM or TGI, or a custom container.
- Hugging Face Hub integrationDownloads model weights from the Hub fast and with the correct security policies.
- Fully managed infrastructureHandles Kubernetes, CUDA versions and VPN configuration so users only deploy their model.
Sources
- Pricing
-
Free Hub access, PRO at $9 a month, Team at $20 and Enterprise at $50 per user, plus hourly compute for Spaces and Inference Endpoints.
- Constraints
-
Official documentation · 2 Oct 2026
Derived from the security facts on this page, from the same source.
- Security
-
Official documentation · 2 Oct 2026
Platform-wide: Hub security doc plus Hub docs pages for SSO, audit logs and storage regions (Hub, Inference Endpoints, Spaces). Caveats: BAA and GDPR DPA are offered through an Enterprise plan; SSO/SCIM/audit logs/EU storage regions are Team/Enterprise plan features; SCIM is Enterprise-plan; Inference Endpoints docs say SOC2 Type 2 covers Endpoints. Training on customer data, ZDR (docs say payloads are not stored, not a named ZDR mode) and encryption at rest not stated, so unknown/absent. Jobs, Chat and Inference Providers left out (not covered by these pages).
- Sold within
-
Its own hourly rate card on the pricing page: "Dedicated inference, starting at $0.033/hour.", followed by per-instance CPU/accelerator/GPU prices. Docs (huggingface.co/docs/inference-endpoints/pricing) add a prerequisite that does not name a plan: "🤗 Inference Endpoints is accessible to Hugging Face accounts with an active subscription and credit card on file."
- Deployment
-
Official documentation · 13 Oct 2026
Page: 'Easily deploy your AI models to production on our fully managed platform'.
- Interfaces
-
Official documentation · 12 Oct 2026
Page: 'using the hf CLI. / Copy prompt / Install hf-cli / Also available via MCP'.
- Modalities
-
Official documentation · 14 Oct 2026
Page: 'text/plain, image/png, image/jpeg, image/jpg, image/tiff, image/bmp, image/gif, image/webp, image/x-image, audio/x-flac, audio/flac'.
- Intended for
-
"Join thousands of developers and teams using Inference Endpoints to deploy their AI models at scale"; dedicated Enterprise plan with custom quote and premium support
Also from Hugging Face
-
Hugging Face Hub Platform
Hosts and versions open models, datasets and demo applications, publicly or privately.
-
Spaces Platform
Hosts runnable demo applications for models on free or paid hardware.
-
HuggingChat Application
Chat app powered by open-source AI models with an Omni router that automatically selects the most suitable model, or lets users pick directly from over 100 open models.
-
Inference Providers API service
Marketplace connecting users to multiple third-party inference providers for hosted AI models, billed per input/output token with provider- and model-specific rates shown side by side.
-
Hugging Face Jobs Infrastructure service
Pay-as-you-go compute for running AI training, fine-tuning, synthetic data generation and batch inference jobs on Hugging Face's own CPU/GPU/TPU infrastructure via CLI or Python API.
-
Storage Buckets Infrastructure service
Object storage for AI teams to hold models, datasets, checkpoints and artifacts, with per-TB pricing, Xet deduplication and a built-in CDN, without git overhead.
-
AutoTrain Platform
AutoTrain provides a no-code interface for training machine learning models.
-
ML Intern AI agent
Mode in HuggingChat that plans and runs machine-learning work from a plain-language request: reads papers, models and datasets, writes scripts, runs them on Hugging Face compute, and can reproduce a paper, finetune and evaluate a model, curate a dataset or deploy a demo Space.
Open Hugging Face in the directory
Something wrong here? Send a correction — quote this product id: huggingface-inference-endpoints.