Hugging Face Inference Providers now route through OpenRouter
Hugging Face Inference Providers now send chat model calls through OpenRouter. Same token, no markup, but your own provider keys stop working.
Hugging Face switched its Inference Providers to OpenRouter's router on November 16. If you call open-weight chat models through Hugging Face, your requests now go through OpenRouter, which picks a provider per call. Your token and code keep working. One thing breaks: you can no longer bring your own provider API key for language models.
What actually happened
Per the Hugging Face announcement and OpenRouter's post, both dated November 16:
- The router changed. Hugging Face replaced its own custom router with OpenRouter's. Each chat request goes to a provider picked on live price, speed and reliability. If that provider fails, the call retries on another one. Before, you got the first provider's error.
- Scope. OpenRouter says 160 models from 38 labs route this way today. Image, video and audio generation still run directly through Hugging Face, not OpenRouter.
- Same API surface. Chat models take OpenAI Chat Completions, OpenAI Responses and Anthropic Messages calls. The
:fastest,:cheapestand named-provider suffixes still work. - Billing. Hugging Face says you pay the provider's price on your Hugging Face bill, with no markup from Hugging Face.
- The breaking change. Language model calls now bill to your Hugging Face account, so custom provider keys no longer work for those models. Image and video provider keys keep working.
Hugging Face calls this the "first step." Next on its list: connecting your OpenRouter account to Hugging Face, and routing more tasks through OpenRouter.
Why the Hugging Face OpenRouter switch matters for your business
Automatic failover is a real upgrade. Most small teams we work with call one provider and eat its outages. Now a capacity error on one host retries somewhere else, with no code change.
But read the fine print. If you set up Hugging Face with your own key for a provider like Together or Novita, maybe for your negotiated rate or your own usage logs, that path is gone for chat models. Your spend now flows through your Hugging Face bill. Check your keys today, before an invoice or a failed call tells you.
Also note OpenRouter's own routing test: fewer capacity errors and faster first tokens, at 6.0% higher cost per request. That was OpenRouter's experiment, not a Hugging Face price change. Still, "best provider" is not always "cheapest provider." If cost matters more than speed, pin :cheapest or a named provider.
The larger pattern: the layer between you and the model keeps getting consolidated. We wrote about this when Stripe and OpenRouter tied up. Keep your model calls behind your own thin interface so a router change is a config edit, not a rewrite.
Key takeaways
- Hugging Face Inference Providers route chat model calls through OpenRouter as of November 16
- OpenRouter says 160 models from 38 labs route this way; image, video and audio skip OpenRouter
- Same token and code; failed calls now retry on another provider
- Custom provider API keys no longer work for language models on Hugging Face
- Hugging Face says no markup on provider prices; pin a provider if cost beats speed
Your model router should be a setting, not a dependency. We build AI features behind a thin interface you own, so swapping Hugging Face, OpenRouter or a direct provider is a one-line change. See how we build, or send us your current model setup for a review.
Sources: Hugging Face, OpenRouter.
- #hugging-face
- #openrouter
- #open-weight-models
- #llm-api
- #inference
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
SnapLogic MCP Discovery Hub: a registry for MCP server sprawl
SnapLogic launched Discovery Hub, a registry of MCP servers with ready configs for Claude Code, Cursor and VS Code. Third-party support comes later.
Read itInfoblox data breach came through a data migration vendor
Infoblox confirmed hackers reached company information through a compromised data migration vendor. Products weren't hit. Revoke migration access when the job ends.
Read it