DeepInfra lands on Hugging Face inference

This is a small integration with a big ergonomics win: it turns DeepInfra from “another inference vendor” into a first-class option inside Hugging Face’s routing layer, which is where a lot of open-model experimentation already starts.

• –HF’s router already supports provider selection by model suffix, so `:deepinfra` fits a familiar workflow instead of forcing a new SDK or endpoint
• –The appeal is strongest for teams that want open-model access without standing up their own inference stack or paying vendor markup on top of per-token pricing
• –DeepInfra’s value prop is breadth and freshness: a large catalog, fast adoption of new open models, and an API surface that stays close to OpenAI-style clients
• –For production users, the real question becomes routing policy: fastest, cheapest, or preferred provider, since those defaults can matter more than the headline price
• –This is infrastructure news, but it also nudges Hugging Face further toward becoming the control plane for open-model inference