Hugging Face Inference Providers via OpenRouter
Hugging Face announced that it is switching its Inference Providers service for language models from its own custom router to OpenRouter's, so requests go through OpenRouter, which picks a provider per call on live price, speed and reliability.12 A Latent Space newsletter item also noted that Inference Providers now runs on OpenRouter with the same endpoint and token.3 Hugging Face states that for every chat model, its router works with the Chat Completions and Responses APIs from OpenAI as well as Anthropic's Messages API. It adds that prices stay as they were: users pay the provider's rate, with no extra Hugging Face charge.45 Image, video and audio generation are not affected and continue to run directly through Hugging Face Inference Providers rather than OpenRouter.6
Drawing on OpenRouter figures, Hugging Face reports that the share of tokens served by open-weight models climbed from 26.8% in October 2025 to 71.7% by the end of September 2026, and that on average these models are roughly four times cheaper than closed ones.78 It also acknowledges that the best open-weight models still trail proprietary ones on the hardest benchmarks, citing the Artificial Analysis Intelligence Index where Kimi K3.1 Pro scores 46 against 65 for Claude Fable 5.2, at roughly a seventh of the cost per task.910
OpenRouter's own measurements show that after it rebuilt its routing, capacity errors fell 46%, time to first token got 16.9% better, and throughput rose 16.3%.11
Users can still steer routing with model-id suffixes such as :fastest, :cheapest or a named provider, but a breaking change means they can no longer use their own provider API keys for language models, since requests are now billed to their Hugging Face account or organization.1213 Hugging Face calls this the first step of the partnership, with plans to let users connect their OpenRouter account and route more tasks through OpenRouter.1415
Sources quoted
Today, we are switching Inference Providers from our own custom router to OpenRouter's.
— Hugging Face, Hugging Face and OpenRouter join forces to make open-weight models easy to runwhich picks the best provider for each call based on live price, speed and reliability, and retries on another provider if the first one fails
— Hugging Face, Hugging Face and OpenRouter join forces to make open-weight models easy to runHugging Face Inference Providers now runs on OpenRouter using the same endpoint and token.
— Latent Space, [AINews] Perplexity hits a $1B revenue run rate, up 4x in 2026The router supports OpenAI's Chat Completions and Responses APIs and Anthropic's Messages API for every chat model, whatever the provider behind it.
— Hugging Face, Hugging Face and OpenRouter join forces to make open-weight models easy to runHugging Face doesn't add a markup on top of it.
— Hugging Face, Hugging Face and OpenRouter join forces to make open-weight models easy to runImage, video, and audio generation continue to work exactly as before, directly through Hugging Face Inference Providers without going through OpenRouter.
— Hugging Face, Hugging Face and OpenRouter join forces to make open-weight models easy to runopen-weight models went from 26.8% of the tokens served in October 2025 to 71.7% at the end of September 2026, after peaking at 77.5% in July
— Hugging Face, Hugging Face and OpenRouter join forces to make open-weight models easy to run- Quote not reprinted: this page already repeats as much of this article as we allow. — Hugging Face, Hugging Face and OpenRouter join forces to make open-weight models easy to run
The best open-weight models still trail the best proprietary ones on the hardest benchmarks.
— Hugging Face, Hugging Face and OpenRouter join forces to make open-weight models easy to run- Quote not reprinted: this page already repeats as much of this article as we allow. — Hugging Face, Hugging Face and OpenRouter join forces to make open-weight models easy to run
- Quote not reprinted: this page already repeats as much of this article as we allow. — Hugging Face, Hugging Face and OpenRouter join forces to make open-weight models easy to run
- Quote not reprinted: this page already repeats as much of this article as we allow. — Hugging Face, Hugging Face and OpenRouter join forces to make open-weight models easy to run
- Quote not reprinted: this page already repeats as much of this article as we allow. — Hugging Face, Hugging Face and OpenRouter join forces to make open-weight models easy to run
This is the first step of our partnership with OpenRouter.
— Hugging Face, Hugging Face and OpenRouter join forces to make open-weight models easy to run- Quote not reprinted: this page already repeats as much of this article as we allow. — Hugging Face, Hugging Face and OpenRouter join forces to make open-weight models easy to run
Who reported it
- Hugging Faceannouncement: Language-model requests now go through OpenRouter, which picks providers by live metrics and retries failures; custom provider keys for language models are no longer supported.
- Latent Spacerepeats Hugging Face
Facts and takes
Facts: 8 checked facts from 3 sources
- Hugging Face Inference Providers: retries a failed request on another provider
- Hugging Face Inference Providers: supports OpenAI Chat Completions and Responses APIs and Anthropic Messages API for every chat model
- Hugging Face Inference Providers: image, video and audio generation without going through OpenRouter
- ERNIE 5.5 Flash Preview is available on OpenRouter
- Hugging Face Inference Providers is available on OpenRouter
- OpenRouter: weekly token volume up 35-fold
- OpenRouter: weekly token volume up 3.5-fold
- OpenRouter: decision-model requests per month 1B
Timeline
- · Hugging Face: Hugging Face and OpenRouter join forces to make open-weight models easy to run
- · Latent Space: [AINews] Perplexity hits a $1B revenue run rate, up 4x in 2026