HFAI Cloud · playground
One prompt.
One model free, three with yours.
Without a key, one prompt goes to gpt-oss:120b. With your own key, the same prompt goes to gpt-oss:120b, deepseek-v4.1-flash and deepseek-v4-pro in parallel, side by side.
01 / THE FREE DEMO
One model without a key.
The anonymous path is deliberately narrow,
and the page says so.
The free tier covers exactly one row of the catalogue: gpt-oss:120b. It is the same model an account calls, served from the same endpoint, but the anonymous path accepts one prompt at a time and returns a short answer. You do not need to sign in, hold a balance or paste a key to use it, and the page will not ask for one.
The budget is shared. A per-address hourly limit and a shared daily ceiling are checked before a request is sent, so a busy day can close the free path early. When that happens the API's own message is printed below the form, word for word, and the page does not retry on your behalf. Nothing on this page is metered against an account, because the anonymous demo has no account to meter.
02 / WHAT IS SENT
Four fields, and a key if you bring one.
The free and keyed requests
share one shape.
- Model. The free demo sends one model, gpt-oss:120b. With your key the same prompt goes to gpt-oss:120b, deepseek-v4.1-flash and deepseek-v4-pro, one request each, sent in parallel.
- Messages. Your prompt, as one user message. There is no system prompt and no history.
- Length. At most 512 tokens come back per model, so each answer is short by design.
- Key. The free request carries no authorization header. A keyed comparison sends your key in the
Authorizationheader of each request and nowhere else: nothing is stored in your browser, and no conversation is kept on our side.
One model without a key; three with yours. A key comparison spends the key owner's credits and is capped at 512 tokens per model. A browser extension with page access can read any key typed into a page, so paste a key only on a machine you trust. The request is a plain chat-completions call, the same shape the API reference documents. A paid key and the wider catalogue are described on pricing.
03 / TRY IT
Send one prompt, or compare three.
One press, one free request.
A key adds two more columns.
With no key, the free demo sends this prompt once to gpt-oss:120b. With a key, three requests go in parallel to gpt-oss:120b, deepseek-v4.1-flash and deepseek-v4-pro. A key comparison spends the key owner's credits and is capped at 512 tokens per model. The key is used only in the request header and is never stored. A shareable link carries the prompt and the three model ids only, is limited to 2,000 characters, and is refused rather than shortened when the prompt will not fit.
deepseek-v4.1-flash
deepseek-v4-pro
04 / WHERE TO GO NEXT
From one answer to the catalogue.
The free demo stops at one model; a key adds two more columns on this page. The model picker indexes all eleven rows by context window and price, without ranking them. The pricing page explains the prepaid balance, and for developers shows the same call from curl, Python and JavaScript with a key of your own.