nemesis

provider checks

Is your model really that model?

Give us a provider address. We ask it and the real model the same questions, and show the difference.

request size →tokens counted+0 tokensyou never sentchecked providerreal model
The same requests to two providers. One of them charges 3 214 tokens more on every request. A real case.
01

The provider under test

The one you have doubts about. The address, the key and the model are its.

02

Reference provider

We compare against the same thing you already trust.

where the reference comes from
known providers

The same model at its official provider. Names differ between providers — pick it from their list.

depth

The key is deleted when the run ends. It never reaches our database or logs.

what we find

01

A different model answers

Cheaper than the one you pay for. We compare the answers with the real model.

02

A hidden instruction on top of your requests

It rides on every request you send. You pay for its tokens.

03

A key resold from a coding tool

Cursor, Claude Code, Kiro and others. That tool’s instruction travels with every request of yours.

04

A wrong token count

More tokens billed than the answer contains. Or the answer comes from a cache instead of being generated.

What proven means here.

Two independent checks gave the same result.

Everything below that bar is shown too, marked for what it is.

There is no leaderboard.

We show measurements for one provider at a time. There is no list.

A dossier opens by its exact address.

look up a provider