DeepInfra on Hugging Face Inference Providers 🔥
Hugging Face
Hugging Face just added DeepInfra as an inference provider on its Hub. Cheaper per-token pricing means easy access to big open models like DeepSeek V4 and Kimi-K2.6.
Based on reporting by Hugging Face — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Hugging Face keeps stacking inference providers onto its Hub, and the newest addition is DeepInfra, a serverless platform known for undercutting most rivals on price per token. It's now live directly on model pages, plugged into both the JS and Python client SDKs, so developers don't have to leave the Hugging Face ecosystem to reach it.
DeepInfra brings a catalog of more than 100 models to the table, covering everything from text generation to image and video generation, plus embeddings. For now, the integration only covers conversational and text-generation workloads, giving users a route to open-weight heavyweights like DeepSeek V4, Kimi-K2.6, and GLM-5.1. Hugging Face says image, video, and embedding support is coming soon, so this is clearly a first phase rather than the full rollout.
The mechanics are familiar to anyone who's used Hugging Face's provider system before. You can bring your own DeepInfra API key and get billed directly by DeepInfra, or route requests through Hugging Face and get billed there instead, with no markup added on top. Model pages now surface which providers support a given model, ranked by whatever order you've set in your account preferences. And if you're working through an agent harness like Pi, OpenCode, or OpenClaw, DeepInfra-hosted models slot in without extra glue code, since Hugging Face's provider layer is already wired into most of those tools.
Practically, using DeepSeek V4 Pro through DeepInfra takes about six lines of code with the OpenAI-compatible client, pointing the base URL at Hugging Face's router and authenticating with an HF token. PRO subscribers get $2 of monthly inference credit to spend across any provider, and free users get a smaller quota to test things out before committing to a paid plan.
My take — AI-written commentary, not fact-checked reporting
Another provider, another price war footnote, and that's exactly the point. Hugging Face isn't building models here, it's building the boring but useful plumbing that lets cheaper open-weight competitors like DeepSeek and Kimi actually reach developers without everyone rolling their own API integration. That's a more durable moat than any single foundation model release, and it's why I'd bet on Hugging Face's aggregator strategy outlasting a dozen flashy model launches this year.
Read more about this at: Hugging Face