A user, HitmanSkar, opened a thread about a 402 error when calling Qwen/QwQ-32B via the Hugging Face JS client, along with a message saying he'd exceeded his monthly Inference Providers credits. According to him, he hadn't "used a drop" of the limit, even spun up a fresh account specifically to test this — and the problem has persisted for at least two months. The thread is split over whether it's an outdated request URL, a bug in the service itself, or the way credits are counted.
HitmanSkar lays out the symptoms in detail: the app gets a 402, and the URL api/inference-proxy/hf-inference/models/Qwen/QwQ-32B/v1/chat/completions opens a "Sorry, we can’t find the page you are looking for" page. The console shows the "You have exceeded your monthly included credits for Inference Providers" message. He insists the problem isn't his code: "I haven’t use a bit" — his stack is JS with hf inference.
John6666 immediately points at the URL and posts a curl example against a different host — router.huggingface.co/hf-inference/models/Qwen/QwQ-32B/v1/chat/completions with an Authorization: Bearer hf_... header. His take: "I think there have been changes to the URL of the request destination." In other words, part of the 402s may stem from hitting a deprecated path rather than any real quota usage.
HitmanSkar pushes back sharply: "404 page error! and I want to use in a React app using js not curl." John6666 then provides a JS version via @huggingface/inference: create new HfInference("hf_..."), then call client.chatCompletion with model, messages, max_tokens, and an explicit provider: "hf-inference". The key point — set the provider manually instead of relying on the default.
A separate thread concerns paid accounts. ribbitribbit365 reports hitting the same error as a PRO user: "But the API requests just all failed since I used up all free credits." He references the promise that PRO users can go beyond free credits with subsequent billing, and notes that switching inference providers changed nothing.
John6666 later confirms: "I encountered the same error today," suggests a fix is in the works, and quotes support: "We’re taking a closer look into this and I’ll update you soon!" In the same reply: "Model does not exist, inference API don’t work." jeyes1997 chimes in briefly: same issue, waiting for a fix.
Later, John6666 drops a link to an announcement by julien-c starting with "Important notice 🚨 For Inference Providers who…" — but the thread contains no details of the announcement itself. alexman83 raises a related issue: he's in an enterprise organization, yet HfApiModel from smolagents burns credits from his personal free account rather than the organization's, even though the docs say otherwise.
The common denominator is simple: people are hitting the 402 and the quota message across very different setups — a brand-new account, PRO, an enterprise organization, the JS SDK, and raw HTTP. Nobody in the thread claims it's purely user-side code, and everyone agrees the api/inference-proxy/... path is broken and requests should go to the router host.
The real point of contention is the nature of the error. One theory: a wrong URL and a missing explicit provider in the request. The other: a billing desync where the wrong credits get charged (personal instead of PRO or organization). There's no consensus, and no official breakdown in the thread — the contents of julien-c's announcement are known only from its title.
router.huggingface.co/hf-inference/models/... to separate a network issue from your app's logic.@huggingface/inference, explicitly set provider: "hf-inference" and verify the call goes to the router host, not api/inference-proxy.HfApiModel) is pulling a token from an env variable tied to your personal account instead of the organization token.Hugging Face Forum — Hugging Face Payment Error 402 & You've Exceeded Monthly Quota
Disclaimer: this is a summary of a discussion, not editor-verified facts. Opinions belong to the participants; the numbers belong to whoever cited them.
Source: discuss.huggingface.co · post in Telegram