Is Your Cheap Claude or GPT API Real?

Point this at any OpenAI-compatible endpoint. Four checks, ~15 seconds, entirely in your browser — your API key never leaves this page. No server of ours ever sees it.

API base URL (e.g. https://api.example.com/v1) API key (stays in your browser — this page makes requests directly to the URL above) Model to test (as the reseller names it)

v1 supports OpenAI-compatible endpoints (/chat/completions), which is what nearly all resellers expose. If the endpoint blocks browser requests (CORS), we'll tell you — that means "not testable from a browser", not "fake". A passing result means the endpoint behaved like the genuine model on this date; it is not a guarantee about future requests.

What the four checks do

Reachability & auth — is the endpoint up, and is your key accepted?
Model identity — the model is asked who created it; the answer is compared to the vendor implied by the model name (claude → Anthropic, gpt → OpenAI, gemini → Google, …).
Response fingerprint — structural fields the genuine provider returns (e.g. system_fingerprint on GPT models) are checked for presence and plausibility.
Latency — time to first token and total round-trip. Informational only.