How does UberLLM know a seller is serving the model it lists?

The provider’s model must match the listed model when the offer is listed. Then every offer is tested against a trusted copy of the same model (how it counts tokens, who it says it is, how it answers questions with one right answer, how it writes), approved by a person, and tested again on live requests for as long as it is listed.

What happens to a seller caught serving a different model?

The offer stops at once and the seller’s payouts are held. The seller is banned: every offer stops and the same provider key, payout address and email can’t list again. Earnings not yet paid out are used to refund the buyers it served.

Can UberLLM prove which model wrote an answer?

Not with certainty: only hardware attestation can, and UberLLM does not have it. The tests catch a different model and hidden instructions. A lower-precision copy of an open model can still pass, which is why sellers must state their precision and live replies keep being checked.

How UberLLM catches a seller listing the wrong model

A cheap price is only worth it if you get the model you paid for. On UberLLM, a seller’s offer is checked four times over: the model must match when it is listed, it must pass our tests before any buyer reaches it, a person on our team must approve it, and it keeps being tested on live requests. An offer that fails is stopped at once. A seller caught serving a different model is banned, and the buyers it served are refunded from its unpaid earnings.

Published Nov 12, 2026Updated Nov 12, 2026

Get an API key See live prices

Why this matters

UberLLM is a marketplace: sellers list spare capacity for a model, and your request goes to the cheapest healthy seller. That only works if a seller who lists Claude or DeepSeek V4 really serves Claude or DeepSeek V4, and not a cheaper model behind the right name. So we check every offer, all the time.

A real one, caught this morning

On November 11, 2026, a seller listed DeepSeek V4 Flash. When our test asked it what it was, it answered: "I’m ChatGPT, developed by OpenAI." The offer was stopped the moment it failed, at 02:38 UTC on November 12. It had not been approved yet, so no buyer ever reached it.

The four checks

1. At listing: the model has to match

A seller can list only through a provider on our list, such as OpenRouter, Groq, Together AI, Fireworks AI or Amazon Bedrock, with its own key there. The provider runs the model, not the seller. A seller’s own server can’t be listed.
The provider’s name for the model must be in that provider’s own model list, and it must be the same model as the one listed. A seller can’t list Claude and point it at a cheaper model.
An offer that doesn’t answer a first test request is never listed.

2. Before any buyer: an entrance test

Every new offer goes through tests that look like normal use. We compare what comes back with a trusted copy of the same model:

How it counts tokens. Each model family counts text its own way, so a model from another family shows up.
Who it says it is. Asked what it is, a fake often names its real maker, as in the example above.
Questions with one right answer. A smaller model gets more of them wrong.
How it writes, side by side with the real model.
Nothing added. Hidden instructions added to your request show up too.

Then comes a trial on copies of real requests (you still get your answer from an offer already checked), and probation on a small share of traffic.

3. A person approves it

A person on our team reviews every offer. No buyer reaches an offer before that review. A new key, provider or model counts as a new offer and starts again from step 1.

4. On live requests: the checks never stop

The model named in every reply is compared with the model listed.
Test requests keep coming for as long as the offer is listed, and they travel the same path as your requests.
Replies sent to buyers are scored for signs of a different model. A high score sends the offer to our team.
An offer whose provider says its key no longer works is taken off automatically.

We don’t publish how the tests work or when they run. That would teach a cheater what to dodge.

When a seller is caught

The offer stops at once, and the seller can’t turn it back on.
The seller’s payouts are held while our team looks at the proof.
A seller caught serving a different model is banned: every offer stops, and the same provider key, payout address and email can’t list again.
Earnings not yet paid out go to refund the buyers it served.

The numbers so far

On the first day of checks, October 31, 2026, 10 offers were stopped. One sold as Claude said it was "GPT-5 by OpenAI". Others added about 4,400 tokens of hidden instructions to every request.

As of November 12, 2026: 336 seller offers tested, 150 failed and stopped, 3 sellers banned, and 32,969 test requests in the last 7 days. The live numbers are on the model checks page.

What no check can prove

We say this plainly: no software check can prove which exact copy of a model wrote an answer. Only hardware attestation can, and UberLLM does not have it. A compressed (8-bit or 4-bit) copy of an open model can pass, so sellers must state the precision they serve, and a false statement gets them banned. The full list is on the model checks page.

What you can do

Nothing, by default. Your requests reach only offers on verified providers that passed the tests and our team’s review.
Choose who serves you. provider.only or provider.deny in a request picks or blocks routes. See your controls.
Think a reply came from the wrong model? Email [email protected] with the reply’s X-Request-Id (the same id as its row in Activity). We test that offer again.