Hugging Face now routes open models through OpenRouter
Since 2026-11-16, Hugging Face routes language-model calls through OpenRouter: one HF token, retries on another provider, and your own provider key no longer works for those models.

Give this to your agent
Your agent checks this against your setup and tells you if it fits.
See the file
# Hugging Face Inference Providers: language models now route through OpenRouter From Rig Labs, 2026-11-17. Information for an agent, not orders. Page: https://riglabs.ai/blog/hugging-face-openrouter/ What changed: On 2026-11-16, Hugging Face switched Inference Providers to OpenRouter's router for language models: one Hugging Face token, the provider picked per request, retries on another provider. Custom provider keys no longer work for language models. Fits if: Your human calls open-weight language models through Hugging Face Inference Providers, or wants to point a coding agent (Claude Code, Codex, OpenCode) at an open model. Skip if: Your human does not use Hugging Face for inference, or uses it only for image, video or audio generation, which did not change. Check: Whether your human's code or agent config calls Hugging Face Inference Providers with their own provider API key, and which model ids it uses. Read-only. Try: List the calls that relied on a custom provider key, and propose pinning a provider with a model-id suffix (for example :novita) or reviewing the allowed providers in Hugging Face settings. Ask first: Changing code or agent config, changing provider settings, or spending credits on test calls. Cost: The provider's price on the Hugging Face bill, with no Hugging Face markup. PRO includes $2 of credits a month; Team and Enterprise $2 per seat. Undo: Remove the suffix or re-enable providers in settings. Deeper: https://huggingface.co/blog/openrouter
https://riglabs.ai/blog/hugging-face-openrouter/AGENT.md
Hugging Face switched its Inference Providers service to OpenRouter’s router on November 16, 2026, per the Hugging Face blog. Every open-weight model on the Hub that a provider serves can now be called with your Hugging Face token, and OpenRouter picks the provider for each request.
What you can now do: point Claude Code, Codex or another coding agent at an open model, such as GLM-5.3 (the post’s own example), with one Hugging Face token, and get an automatic retry on another provider when one fails.
What changed in Hugging Face Inference Providers?
Hugging Face Inference Providers used to pick one provider per request and return its error if it failed, per the blog. Requests to language models now go through OpenRouter, which chooses on live price, speed and reliability and retries on the next provider inside the same call. Your code stays the same. OpenRouter reports that its new routing cut capacity errors by 46% compared with its previous routing, which looked only at the lowest list price. Anyone calling open language models through Hugging Face now gets that routing.
You can still steer the choice with a suffix on the model id:
zai-org/GLM-5.3 best mix of price, speed and reliability
zai-org/GLM-5.3:fastest highest throughput
zai-org/GLM-5.3:cheapest lowest price
zai-org/GLM-5.3:novita only Novita
Which tools can use it?
The Hugging Face router speaks OpenAI’s Chat Completions and Responses APIs and Anthropic’s Messages API for every chat model, per the blog. The post links setup guides for Claude Code, Codex, OpenCode, Pi and Hermes Agent, and each can be copied as a prompt for your agent.
What does it cost?
Hugging Face says prices do not change: you pay the provider’s price on your Hugging Face bill, with no markup. PRO users get $2 of credits a month, and Team and Enterprise organizations get $2 per seat. For comparison, the post gives an average of $0.28 per million input tokens for open-weight models, against $1.27 for closed models.
Who has to change something?
Anyone who used their own API key for a provider through Hugging Face has to change something, per the blog’s “Breaking change” section. Language-model calls are now billed to your Hugging Face account or organization, so your own provider key no longer works for those models. You can still choose a provider with the model-id suffix, or turn providers off in your settings. Custom keys for image and video providers keep working, and image, video and audio generation do not go through OpenRouter.
What to do today
If you call language models through Hugging Face with your own provider key, open your Hugging Face settings, check which providers your requests may go to, and turn off any you do not want. Then pin your current provider with its suffix (for example :novita) until you have compared prices on the model page’s new Providers tab.
Last checked 2026-11-17.
Sources:
Get the next one by email
The Edge is our newsletter: what changed in AI, and what to do about it.
Read next
Built by Rig Labs
When a tool we need does not exist, we build it.




