API relay quality test
Find risky API relays before they reach production.
Test whether an OpenAI-compatible relay can list models, complete chat requests, stream correctly, return usable output, and route to the model family you expected.
Catch the problems HTTP checks miss
A relay can return HTTP 200 while producing empty output, broken stream chunks, or answers from a different model family. The result view makes those failures visible.
What Relay Quality Means
A good relay should expose OpenAI-style endpoints, route requests to the expected model family, support streaming chunks, avoid empty output, and keep TTFT and output speed within a usable range.
Relay Problems This Test Looks For
The test can surface relays that list unavailable models, return HTTP 200 with no useful answer, stream too slowly, fail /v1/models, or route a claimed Gemini, Claude, GPT, or DeepSeek model to another family.
Who Should Use It
Use it before buying relay credits, switching a production app to a new gateway, comparing proxy providers, or sharing an OpenAI-compatible endpoint with a team.
FAQ
Is an API relay the same as an official model provider?
No. A relay usually forwards requests to one or more upstream model providers through an OpenAI-compatible interface.
Why can a relay pass HTTP checks but fail quality checks?
HTTP status only shows that the server responded. It does not prove streaming, output speed, model routing, or response content quality.
Can this test prove a model is fake?
No single prompt can prove that absolutely, but mismatched model-family answers are useful risk signals for further checking.
Should I test both streaming and non-streaming requests?
Yes. Some relays handle normal chat completions but break or delay streaming responses.
Do not buy relay credits blind.
Run a quick relay check before switching a real app or sharing an endpoint with your team.