Hugging Face and OpenRouter: open models with one account, what changes

On November 16, 2026, Hugging Face announced a deal with OpenRouter. From now on, requests to language models that go through Inference Providers, the Hugging Face service for using models in the cloud without installing anything, are routed by OpenRouter's system. With a single Hugging Face token you can call hundreds of open models, hosted by dozens of different companies.

If you already write code on Hugging Face, the company says everything keeps working. With one exception, which we cover below.

What do Hugging Face and OpenRouter do together?

Hugging Face is the big archive of open models, the ones with downloadable "weights". OpenRouter works like a switchboard: it takes the request and sends it to one of the providers running that model on their servers. Hugging Face used to have its own router; now it uses OpenRouter's.

For each call the system picks the provider based on current price, speed and reliability. If the first one doesn't answer, the request moves to another one on its own. If you want to decide, you add a suffix to the model name:

  • :fastest picks the fastest provider;
  • :cheapest picks the one that costs least;
  • a provider name, for example :novita, locks the choice to that provider.

The service accepts the API formats of OpenAI (Chat Completions and Responses) and Anthropic (Messages). Hugging Face has published guides for connecting it to coding assistants such as Claude Code, Codex and OpenCode. Images, video and audio stay out: those requests still go straight to the Inference Providers partners.

How much does it cost?

Hugging Face adds no markup: you pay the provider's price, which shows up on your Hugging Face bill. This is what the pricing page says is included:

AccountIncluded monthly creditsBeyond the credits
FreeNoneBuy credits
PRO2 dollarsPay as you go
Team and Enterprise2 dollars per seatPay as you go

Two dollars a month is enough to try things, not to work with every day. The real comparison is the price per token (a token is a small piece of a word). In the announcement Hugging Face quotes OpenRouter's figures: open models cost on average 0.28 dollars per million input tokens and 1.62 per million output tokens, against 1.27 and 6.36 for closed models.

Are open models as good as closed ones?

Not yet, at least according to the numbers Hugging Face quotes. On the Artificial Analysis Intelligence Index, an independent ranking, Kimi K3.1 Pro (open) scores 46 and Claude Fable 5.2 (closed) scores 65. The cost per task, though, is 3.68 dollars against 26.54. For a lot of everyday work, like summaries, drafts and pulling data out of documents, that quality gap matters less than the price gap.

Companies are noticing. Again according to OpenRouter, the share of tokens handled by open models on its platform went from 26.8% in October 2025 to 71.7% in September 2026. If you want the names, you'll find them in our guide to Chinese open-source models.

What changes for people who work with AI?

If you only use AI from a chat window in the browser, nothing. The news matters if you connect models to spreadsheets, automations or internal tools, or if you have to keep an eye on spending at work. One account, one bill and the option to switch model by changing one line are real conveniences.

One change can break things. If you had added your own API keys from language-model providers to Hugging Face, you can no longer use them: those requests are now billed through your Hugging Face account. Personal keys for image and video providers still work. If you have automations built this way, check them before they stop answering.

Then there's privacy. The text you send goes through Hugging Face and OpenRouter and reaches a provider you often don't choose. Organizations can exclude the providers they don't want, and those settings apply to every request billed to the company. We would use it to test different models on non-confidential material. For contracts and customer data, a model running on your own computer is the safer bet: we cover it in Local AI: what it is, when it's worth it and what you need.

Frequently asked questions

Do I need an OpenRouter account?

No, a Hugging Face token is enough. Hugging Face says the next step will be letting you connect an existing OpenRouter account, with no date given.

Do I have to rewrite my code?

According to Hugging Face, no: address, suffixes and settings stay the same. The only case to fix is personal API keys for language models.

Is the service faster than before?

OpenRouter states, based on its own measurements, 46% fewer capacity errors, a 16.9% shorter time to the first word and 16.3% higher response speed. These are the company's figures and nobody else has checked them yet.

Official sources

This article was written with the help of artificial intelligence models, starting from the official sources listed, and is reviewed before publication.

Condividi questo articolo sui tuoi socialShare this article on your social networksComparte este artículo en tus redes socialesTeile diesen Artikel in deinen sozialen Netzwerken

No products were found matching your selection.

Iscriviti alla newsletter di You vs AISubscribe to the You vs AI newsletterSuscríbete al boletín de You vs AIAbonniere den Newsletter von You vs AI