元初AI

Grade C means the site did not meet the B threshold: uptime below 97%, latency above 6.5 seconds, or too little data.

The risk flag is derived from the grade and the sample. Watch — grade B, grade A with a thin sample, or grade C with uptime at or above 95%.

The availability score is 0.75 × uptime + 0.25 × max(0, 100 − latency × 8). It orders sites inside the same window only. A profile with 170 samples carries less confidence than one with thousands; below 200 samples we say watch, not recommended.

Model families recorded on this site: OpenAI GPT. The table lists model IDs that the coverage map matched to this host. A missing row means no match, not that the site lacks the model.

Payments recorded: WeChat Pay, Alipay, Bank transfer, Credit card. Refund policy recorded: yes. Invoice support recorded: yes. Enterprise verification.

These fields come from the public directory snapshot, not from a contract review. Confirm them on the site before paying.

• It does not claim the site is safe, honest, or compliant. We do not audit corporate ownership, fund custody, data handling, or licences.
• It does not claim the models are genuine. Model authenticity requires the seven checks; availability does not answer it.
• It does not claim 24/7 monitoring. A probe window is a sample, and results can change by time of day.
• It is not a paid placement. Certification covers verification and retesting; it does not change the grade or the order of this directory.

Is 元初AI reliable?

We publish what we measured: 99.6% uptime and 9.03s average latency across 170 samples in the 7-day window. That is an availability observation, not a recommendation. We do not audit ownership, fund custody, data handling, or licensing.

What does grade C mean?

Grade C means the site did not meet the B threshold: uptime below 97%, latency above 6.5 seconds, or too little data. Detection failures are not folded into this grade yet.

Does 元初AI serve the models it advertises?

The directory records advertised model IDs. To test whether a relay serves the genuine model, run the seven checks with a key you control. A single run is a spot check; results can change by time of day.

How do I test 元初AI myself?

Open the detection page, enter the site's base URL and a key, and run the checks. The key is used for the request and is not stored. You can also compare this profile with other sites in the directory.