AI 中转降智检测
测测你正在用的中转 API 是不是满血模型 —— 花高配的钱,别被偷偷换成便宜小模型。填你自己的中转信息,一键跑 6 道探针。
How it works
Six deterministic probes with known-correct answers — base64 decoding, long-context needle recall, reasoning traps, instruction following, refusal behaviour and output-format compliance. A full-strength model passes nearly all of them; a silently downgraded or swapped model fails a predictable subset. You get a 0-100 score plus a per-probe breakdown, in about 10-30 seconds.
FAQ
Some relays quietly route your request to a cheaper, smaller model than the one you asked for — you pay flagship prices and get budget output. Others trim context windows or strip capabilities. This tool detects both.
It sends six deterministic probes whose correct answers are known in advance (base64 decoding, needle-in-a-haystack recall, reasoning traps, instruction following). A full model passes almost all of them; a downgraded one fails a predictable subset. You get a 0-100 score and a per-probe breakdown.
The key is used for that one test run and is never stored, logged or reused. The whole methodology is open source, so you can read exactly what happens — or run the CLI locally instead.
Yes — that is the point. It works against any OpenAI-compatible endpoint. Test your current provider, test us, compare. We publish our own score rather than asking you to trust us.