RunAPI

Grade B means at least 97% measured uptime and average latency of 6.5 seconds or less in the 7-day window.

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 813 samples carries less confidence than one with thousands; below 200 samples we say watch, not recommended.

Model families recorded on this site: Anthropic Claude, DeepSeek, Google Gemini, Zhipu GLM, OpenAI GPT, xAI Grok, Moonshot Kimi. 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.

Showing the first 40 of 92 model IDs linked to this host in the coverage snapshot.

Payments recorded: 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 RunAPI reliable?

We publish what we measured: 98.0% uptime and 6.18s average latency across 813 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 B mean?

Grade B means at least 97% measured uptime and average latency of 6.5 seconds or less in the 7-day window. Detection failures are not folded into this grade yet.

Does RunAPI 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 RunAPI 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.