OpenAI API Checker

It checks model listing, selected model availability, OpenAI-compatible response shape, streaming behavior, latency, and a fixed set of control questions for instruction following, basic knowledge, structured output, and model-family behavior.

Latency is the time from sending a normal request to getting the first response. TTFT, or first token time, is when the first streamed word or chunk appears. Streaming total is the full end-to-end streaming time. Tokens/sec is output speed. Input tokens are the prompt tokens you sent, and output tokens are the tokens the model returned.

The score combines multiple technical signals instead of trusting one field. That makes it easier to notice relays that look usable at first but have unstable structure, weak model behavior, or slow response patterns.

No. The API key is used only for this live test. It is not saved to an account system, ranking database, or hosted-key feature. Prefer a low-scope or temporary test key.

No. Treat it as a technical health check before buying, integrating, or relying on a relay. Long-term stability and business risk still require repeated testing and your own judgment.

The score is a single-run quick-check score built from question accuracy, model-target match, protocol and response-shape consistency, plus streaming and latency signals. It is not a long-term benchmark or statistical confidence score.

Relays may switch upstream providers, routes, or regions. Latency and response behavior can also fluctuate, so repeated tests are more meaningful than a single result.