Hugging Face Integrations & Stack Fit 2026
Hugging Face Inference Providers is the hub for open models: one token reaches 200+ models across 19 providers behind one OpenAI-compatible API, with zero lock-in.
Hugging Face Integrations verdict
Hugging Face Inference Providers is the hub layer for open models.
A single Hugging Face token reaches more than 200 models across 19 partner providers, including Groq, Together, Replicate, SambaNova and Fireworks, behind one OpenAI-compatible API at https://router.huggingface.co/v1. It integrates four ways.
Choose Hugging Face Inference Providers when you want open-model access with provider flexibility from one integration. Keep your OpenAI SDK and point it at https://router.huggingface.co/v1 with an HF token for chat. For image, embeddings or speech in the same app, use the huggingface_hub or @huggingface/inference clients. Let provider='auto' pick the fastest host with automatic failover, add :cheapest when cost matters, or pin a provider when you need consistent behaviour. For teams, route spend through X-HF-Bill-To, set a spending limit and disable providers you don't want. Prototype in the Playground or the model-page widget, then lift the generated snippet into code. The trade-off is that this is a routing and access layer over open models. Model quality is the model's, the OpenAI-compatible path is chat-only, and heavy production users move past the small free credits into pay-as-you-go.
- The OpenAI-compatible router handles chat completions only. For image, embeddings or speech you use the huggingface_hub or huggingface.js clients, not a base-URL swap.
- Provider coverage is per-task. The partner matrix shows which of the 19 providers serve chat, embeddings, image, video or speech, so not every model runs everywhere.
- A Custom Provider Key changes the billing model. HF monthly credits do not apply and the provider bills you directly, in exchange for provider-specific features.
- Models
- 200+ across 19 providers
- OpenAI compat
- router.huggingface.co/v1
- Clients
- huggingface_hub, huggingface.js
- Agents
- Codex, Claude Code, Copilot
- Lock-in
- Zero, switch providers
This page covers how Hugging Face fits a stack. The raw model quality and real cost live on their own pages.
Check your stack fit with Hugging Face
- OpenAI-compatible router, API, Chat completions (Drop-in OpenAI base URL, Server-side provider selection).
- Auth: HF token (Bearer).
- Programmatic access: Org Billing (X-HF-Bill-To header), Model Discovery (GET /v1/models), Openai Compatible Router (router.huggingface.co/v1).
Based on Hugging Face's documented connector matrix. No native connector means API or CSV is the path.
Where Hugging Face fits in a stack
- Inference Providers gives developers access to hundreds of machine-learning models powered by partner inference providers, integrated into HF's JS and Python client SDKs for serverless inference.
- It is a drop-in replacement for the OpenAI chat completions API, so existing OpenAI-based code migrates by swapping the base URL to https://router.huggingface.co/v1 and the key.
- It spans 19 partner providers (Cerebras, Cohere, DeepInfra, Fal, Featherless, Fireworks, Groq, HF Inference, Hyperbolic, Novita, Nscale, OVHcloud, Public AI, Replicate, SambaNova, Scaleway, Together, WaveSpeedAI, Z.ai) through one consistent API.
- There is zero vendor lock-in: you get models from Cerebras, Groq, Together AI, Replicate and more through one interface rather than a single provider's catalog.
- Coding agents plug in directly: a single Hugging Face token points OpenCode, Pi, Codex, Claude Code or the Hermes Agent at the latest open models.
- Hub features run on the same plumbing: Inference Widgets on model pages, the Inference Playground and Data Studio AI all use Inference Providers, so testing and production share one path.
- It is a single all-in-one API across text generation, image generation, embeddings, NER, summarization and image classification, so one integration covers many tasks.
Hugging Face native integrations
| Integration | Type | Capabilities | Setup |
|---|---|---|---|
| OpenAI-compatible router | API | Drop-in OpenAI base URL · router.huggingface.co/v1 · Server-side provider selection | — |
| huggingface_hub (Python) | SDK | InferenceClient · Automatic provider routing · chat / image / embeddings / speech | — |
| @huggingface/inference (JS/TS) | SDK | InferenceClient · TypeScript support · Automatic provider selection | — |
| Coding agents (Codex, Claude Code, OpenCode, Pi, Hermes) | Agent setup | Point agent at Inference Providers · Latest open models · Per-tool setup guides | — |
| VS Code + GitHub Copilot | IDE | Open models in Copilot · Documented integration guide | — |
| Inference Widgets | Hub UI | Interactive model-page widget · Copy code snippets · Entry point to test a model | — |
| Inference Playground | Hub UI | Chat across models and providers · Compare responses · huggingface.co/playground | — |
| Data Studio AI | Hub feature | Converts text to SQL on dataset pages | — |
| Custom Provider Key | Billing integration | Bring your own provider account · Direct provider billing · Provider-specific features | — |
API, webhooks and automation
| Mechanism | How | What it does |
|---|---|---|
| Automatic routing | provider="auto" (default) | Picks the fastest available provider, with failover if one is down |
| Selection policy | :fastest / :cheapest / :preferred | Append a policy suffix to the model id to steer cost vs speed |
| Provider pinning | model:provider (e.g. :sambanova) | Force a specific provider for a request |
| Org billing header | X-HF-Bill-To: my-org-name | Attribute inference cost to a Team/Enterprise org or resource group |
| Model discovery | GET /v1/models | List models across providers with pricing, context, latency, throughput |
| Hub provider filter | ?inference_provider=fireworks-ai | all | Filter Hub models by the provider that serves them |
Common Hugging Face stack recipes
- Migrate existing OpenAI chat code with minimal changes: keep the OpenAI SDK, set baseURL to https://router.huggingface.co/v1 and the key to your HF token (chat completions only).
- Build a multi-task app on one client: the InferenceClient covers text generation, text-to-image, embeddings and speech, so one token and one library serve many tasks.
- Wire a coding agent: point Codex, Claude Code, OpenCode, Pi or the Hermes Agent at Inference Providers with a single Hugging Face token to run the latest open models.
- Consolidate team billing: each member keeps their own token while requests bill to the org via X-HF-Bill-To, and admins set a spending limit and disable specific providers.
- Bring your own provider key when you already have a provider account and want provider-specific features or direct billing; set the key in HF settings and your code stays unchanged.
- Prototype before code: test a model in the Inference Playground or the model-page widget, then click 'View Code Snippets' to drop the generated Python or TypeScript into your app.
Hugging Face integration limits and auth
- Auth needs a fine-grained Hugging Face token with 'Make calls to Inference Providers' permission, created in token settings.
- Monthly free credits are small: $0.10 for free users, $2.00 for PRO and $2.00 per seat for Team/Enterprise, after which continued use requires purchasing credits.
- Monthly credits do not apply to the Custom Provider Key path: with your own key the provider bills you directly and HF credits are not used.
- The OpenAI-compatible endpoint is for chat completion only; other tasks (text-to-image, embeddings, speech) must use the Hugging Face inference clients.
- Provider coverage is per-task, not uniform: the partner table shows which providers support chat (LLM/VLM), feature extraction, text-to-image, text-to-video or speech-to-text, so not every model runs on every provider.
- As of July 2025 the built-in hf-inference provider focuses mostly on CPU inference (embeddings, ranking, classification, or smaller historical LLMs like BERT or GPT-2), with compute-time billing past the free tier.
Hugging Face Integrations FAQ
How does Hugging Face integrate with my stack?
Four ways, all on one Hugging Face token. Use the OpenAI-compatible router at https://router.huggingface.co/v1 to drop into existing OpenAI code, for chat only. Use the native huggingface_hub Python or @huggingface/inference JS clients for all tasks, including image, embeddings and speech. Point a coding agent, like Codex, Claude Code, OpenCode or VS Code with Copilot, at Inference Providers with a single token. Or test in the Hub's Inference Widgets and Playground and copy the generated snippet.
Which providers and models can I reach through Hugging Face?
More than 200 models served by 19 partner providers, including Cerebras, Cohere, DeepInfra, Fireworks, Groq, Novita, Replicate, SambaNova, Together and Z.ai, all behind one consistent API. Coverage is per task. The partner matrix shows which providers handle chat, feature extraction, text-to-image, text-to-video or speech-to-text.
Can I control which provider runs my request?
Yes. By default provider='auto' picks the fastest available provider and fails over if one is down. Append a policy suffix to the model id to change that, :cheapest for lowest price or :preferred for your order, or pin a provider explicitly with model:provider, as in openai/gpt-oss-120b:sambanova. GET /v1/models lists models across providers with per-provider pricing, context, latency and throughput.
How is billing handled across providers?
Centrally. Requests routed by Hugging Face bill to your HF account at the provider's rate with no markup, drawing on monthly credits, $0.10 free, $2 PRO or $2 a seat, before pay-as-you-go. Teams attribute spend with the X-HF-Bill-To header and can set spending limits or disable providers. Alternatively, a Custom Provider Key bills you directly through the provider, in which case HF credits do not apply.
Do I need a separate account for each provider?
No. The default routed path means Hugging Face handles provider accounts and billing for you, so one HF token reaches all 19 providers. You only need a provider account if you choose the Custom Provider Key path, to use a provider's own billing or specific features. You set that once in your Hub settings while your code stays unchanged.
Sources & verification
| Source | What was checked | Last checked |
|---|---|---|
| Hugging Face Official | Official product page | July 10, 2026 |
| Hugging Face | Inference Providers | July 10, 2026 |
| Hugging Face | Inference Providers Hub Integration | July 10, 2026 |
| Hugging Face | Pricing and plans | July 10, 2026 |
| Hugging Face | Inference Providers First API Call | July 10, 2026 |
Every fact on this Hugging Face page is tied to a named source and a verification date. Freshness-sensitive figures trace to the sources above; verify against the vendor before relying on them.
Explore Hugging Face
Every page on Hugging Face in one place, you are on integrations.
Snapshot, score and verdict
How to get API access, limits, SDKs and what it costs
You are here
Every tier and the entry price
Compared and ranked vs peers
Price and feature change history
Browse the full Large Language Models category