Hugging Face switched its Inference Providers service from its own router to OpenRouter’s on 16 November. Calls to language models made with a Hugging Face token now go through OpenRouter, which picks a provider for each request. Hugging Face says your code stays the same.
What changes for callers
- Retries. Before, Inference Providers picked one provider and returned its error if that provider failed. Now a failed call is retried on the next provider inside the same request.
- Routing by suffix. The default balances price, speed and reliability. Add
:fastestor:cheapestto the model id, or a provider name such as:novitato pin one. - Prompt caching. Routing tries to keep a conversation on the provider holding its cache. To make that explicit, send a stable UUID per conversation in the
X-HF-Session-Idheader. - APIs. Every chat model is reachable through OpenAI’s Chat Completions and Responses APIs and Anthropic’s Messages API, whatever the provider behind it. Hugging Face has setup guides for Claude Code, Codex, OpenCode, Pi and Hermes Agent.
Hugging Face quotes OpenRouter’s own measurements for the new routing: 46% fewer capacity errors, 16.9% faster time to first token and 16.3% more throughput.
The breaking change
Because language-model requests are now billed to your Hugging Face account or organisation, you can no longer use your own API key for a provider with these models. You can still pin a provider with the suffix, or switch providers off in your user or organisation settings so the router never sends traffic there. Custom keys for image and video providers keep working, and image, video and audio generation do not go through OpenRouter at all.
Prices and visibility
Hugging Face says prices don’t change: you pay the provider’s price on your Hugging Face bill, with no markup. PRO accounts get $2 of credits a month, and Team and Enterprise organisations get $2 per seat. Model pages gain Providers and Metrics tabs with each provider’s price, speed and uptime over the last 24 hours, and the usage overview splits spend by provider.
The post does not say how data handling changes now that OpenRouter sits in the path. Zero-data-retention options on OpenRouter depend on the provider and model, as our look at Google models and ZDR routing showed, so teams with retention rules should check the provider list for each model they use. That includes popular open-weight models such as DeepSeek V4.1 Flash 1017, which Hugging Face says is this week’s most used model on OpenRouter.
Why this matters
If you call open-weight chat models with a Hugging Face token, you get automatic failover and simple speed or price routing without changing code. If you relied on your own provider key for a chat model, that setup stopped working and needs to move to a pinned provider or to the provider directly. And with a second company now in the request path, review which providers are switched on for your account before sending sensitive data.

The Campfire
CommentsThanks. Your comment is waiting for the desk. Once it is approved it appears here within about a minute.
Add a comment